Best Practices of Generative AI Across Industries in China by 2024 PDF Free Download
1 / 100/100
100%
Best Practices of
Generative AI Across Industries in
China by 2024
AIGC, Content Innovation, Cross-Border Integration, Knowledge Graph, Data Governance
Leadleo Research Institute
Frost & Sullivan
August 2024
Description
Sullivan, in conjunction with Leadleo Research Institute, would like to release
China Generative AI Industry Best Application Practices 2024, a series of
reports on Generative AI in China. This report aims to sort out the
development trend of generative AI technology, clarify the demand for
generative AI in various industries, and screen out the best application
practices of generative AI in various industries based on complete selection
indexes and processes.
Sullivan, in collaboration with Leadleo, has conducted research on Generative
AI vendors and the downstream enterprise customers they work with. The
current development status of Generative AI in various industries provided in
this market report also reflects the overall development trend of Generative
AI. The report's selection of best practices is based on the cases submitted by
vendors to Sullivan, and applies only to this year's development cycle of
generative AI cases in China.
All figures, tables, and data in the text of this report are derived from surveys
conducted by Frost & Sullivan Consulting (China) and Leadleo Research
Institute, and the data are rounded to one decimal place.
Any content provided in the report (including but not limited to data, text, charts,
images, etc..) is a highly confidential document exclusive to Frost & Sullivan and Leadleo
Institute (unless otherwise indicated in the report). Without the prior written permission
of Frost & Sullivan and Leadleo Institute, no one is allowed to copy, reproduce,
disseminate, publish, quote, adapt or compile the contents of this report in any way
without authorization, and Frost & Sullivan and The Leadleo Institute reserve the right to
take legal measures and pursue the responsibility of the relevant persons in the event of
any violation of the above agreement. All commercial activities carried out by Frost &
Sullivan and The Leadleo Institute use “Frost & Sullivan”, Sullivan”, Leadleo
Institute”or“Leadleo”.“The Frost & Sullivan and Leadleo Institutes do not have any
affiliates other than those listed above, nor do they authorize or employ any other third
party to conduct business on behalf of Frost & Sullivan or Leadleo Institutes.
www.leadleo.com
400-072-5588
Frost & Sullivan Market Insight
3
◆Introductory ---------- 5
•Definition and Scope ---------- 6
•Background and Purpose ---------- 7
•Selection Process and Methodology ---------- 8
◆Scoring Dimension Analysis ---------- 9
•Functionality and Adaptability Dimension ---------- 10
•Performance and Innovation Dimension ---------- 11
•Deployment and Support Dimension ---------- 12
•Experience and Satisfaction Feedback Dimension ---------- 13
◆Compendium of Best Practices of Generative AI Industry ---------- 14
•Best Practices of Generative AI Collection and Analysis ---------- 15
•State of Generative AI Core Technology Applications ---------- 16
•Comprehensive Atlas of Generative AI Industry Best Practices ---------- 17
•Challenges and Developments in The Gaming and Entertainment
Industry ---------- 18
•Potential Application Risks in The Gaming and Entertainment
Industry ---------- 21
•Best Practices in The Gaming and Entertainment Industry ---------- 22
•Challenges and Developments in The Manufacturing Industry ---------- 26
•Potential Application Risks in The Manufacturing Industry ---------- 29
•Best Practices in The Manufacturing Industry ---------- 30
•Challenges and Developments in The Healthcare Industry ---------- 34
•Potential Application Risks in The Healthcare Industry ---------- 37
•Best Practices in The Healthcare Industry ---------- 38
•Challenges and Developments in The Financial Industry ---------- 42
•Potential Application Risks in The Financial Industry ---------- 45
•Best Practices in The Financial Industry ---------- 46
•Challenges and Developments in The ICT Industry ---------- 50
•Potential Application Risks in The ICT Industry ---------- 53
•Best Practices in The ICT Industry ---------- 54
Research Framework (1/2)
www.leadleo.com
400-072-5588
Frost & Sullivan Market Insight
4
•Challenges and Developments in The Public Service Industry ---------- 58
•Potential Application Risks in The Public Service Industry ---------- 61
•Best Practices in The Public Service Industry ---------- 62
•Challenges and Developments in The Automotive Industry ---------- 66
•Potential Application Risks in The Automotive Industry ---------- 69
•Best Practices in The Automotive Industry ---------- 70
•Challenges and Developments in The Consumer and Retail Industry ---------- 74
•Potential Application Risks in The Consumer and Retail Industry ---------- 77
•Best Practices in The Consumer and Retail Industry ---------- 78
•Challenges and Developments in The Education Industry ---------- 82
•Potential Application Risks in The Education Industry ---------- 85
•Best Practices in The Education Industry ---------- 86
•Challenges and Developments in The Enterprise Application
Industry ---------- 90
•Potential Application Risks in in The Enterprise Application Industry ---------- 93
•Best Practices in The Enterprise Application Industry ---------- 94
◆Methodology ---------- 98
◆Legal Notices ---------- 99
Research Framework (2/2)
•Through algorithms that simulate the human brain and decision-making process, generative
AI can respond to user needs with newly generated content, releasing "cognitive surplus"
opportunities for users and significantly reducing the cost and threshold of content creation.
The operation of generative AI is a continuous iterative process, through continuous model
readjustment and evaluation, so as to generate higher quality and more accurate content.
•As generative AI formally moves into a rapid development phase, generative AI tools
empowering various industries are blossoming, assisting enterprises to enhance strategic
benefits such as reducing costs and increasing efficiency, accelerating innovation, and so on.
This report integrates leading cases from various industry scenarios to provide enterprises
and the market with a showcase of technical capabilities, inspiration for innovative thinking,
and popularization of generative AI-enabling approaches.
•The best case selection process is vendor registration and participation, case input, case
assignment, and final case delivery and release. The case selection will combine industry
cross big data with Sullivan's innovative all-dimensional selection metrics to objectively and
fairly present the best cases in the generative AI industry.
❑0.1 Definitions and Scope
❑0.2 Background and Purpose of The Report
❑0.3 Best Practices Selection Process and Methodology
Introductory
6
www.leadleo.com
400-072-5588
Key findings
Through algorithms that simulate the human brain and decision-making process, generative AI can
respond to user needs with newly generated content, releasing “cognitive surplus”opportunities for
users and significantly reducing the cost and threshold of content creation. The operation of generative
AI is a continuous iterative process through continuous model readjustment and evaluation, so as to
generate higher quality and more accurate content.
0.1 Definitions and Scope
❑Generative AI starts with sophisticated deep learning
models that can be used to create new ideas and
content in response to user prompts and requests.
Source: Sullivan
Generative AI Industry Definition
Generative AI Operational Process and Scoping
Generative
AI
AI
Machine
Learning
Deep LearningTraining models
based on raw data
Error feedback
and self-learning
Generate new
content
Demand
Manage
-ment
Input
Filtering
&Model
I/O
Enginee
ring
Model Zoo (GAN,
LLM, etc..)
Model Selection
Telemetry: observability, monitoring
Output
Evaluati
on
Orchestration
Output
Text Imagery
Video Audio
Code 3
D
TO C
TO B
TO P
The scope
of the
report
focuses on
To B
enterprise
service
cases
Fine Tuning, RAG, Prompt Engineering
❑Generative AI operates as a continuous iterative process that requires constant model tuning and
evaluation of the generated results to arrive at better quality generated results.
Sullivan Market Research
Data Management Users
The operation process of generative AI contains four main phases: model training, model selection, content generation evaluation
and model retuning, of which model retuning is an important step for generative AI model updating, accuracy and relevance
improvement.
Introductory: Industry & Research Introduction
Generative AI analyzes and encodes structures and
patterns in large amounts of existing data to process a
user's natural language request and generate new content
in response. Generative AI typically requires larger
datasets for training to improve its ability to generate
diverse content. However, specific data requirements may
vary depending on the task and model architecture. These
AI systems can be used to create new and meaningful
content, including images, video, code, audio, and more.
7
www.leadleo.com
400-072-5588
Key findings
As generative AI formally moves into a rapid development phase, generative AI tools spring up, assisting
enterprises to enhance strategic benefits such as reducing costs and increasing efficiency, accelerating
innovation, and so on. This report integrates leading cases from various industry scenarios to provide
enterprises and the market with demonstrations of technical capabilities, inspiration for innovative
thinking, and popularization of generative AI-enabling approaches.
Background and Purpose
Source: Sullivan
❑Background of the report: Generative AI has entered a rapid development phase.
❑Purpose of the report: To provide industries actively exploring generative AI with demonstrations
of technical capabilities, inspirations for innovative thinking, and ways for generative AI to
empower industries through the presentation of the best case studies.
1Demonstration of technical capabilities: The technical dimension assessment in the excellent cases intuitively
demonstrates the core technical capabilities and innovative performance of the generative AI service solutions to
various industries, providing significant and practical value benefits for enterprise practices.
2Inspiration of innovative thinking: the report integrates leading cases from multiple industries and modes to
help enterprises with the intention of deploying generative AI to explore the integration of various scenarios with
generative AI and to inspire innovative thinking on generative AI functional scenarios in various industries.
3Popularization of Generative AI Empowerment Methods: By demonstrating how Generative AI assists
enterprises in solving their actual pain points, it helps industries better understand the principles and
applications of Generative AI and enhances the market's attention to Generative AI.
Automotive
Public
Service
Healthcare
ICT
Generative
AI
Consumer and Retail
Gaming and
Entertainment
Enterprise Application
Education
Financial
Manufacturing
Improve Process
Efficiency Tapping into Market
Opportunities
Accelerate
Innovation
Improve the
Customer Experience
0.2 Background and Purpose
Since 2012, when deep learning algorithms made major breakthroughs in speech and image recognition, generative
AI has begun to enter a period of rapid development. Today, coupled with the macro environment of accelerating
enterprise digitization, enterprises in various industries have started to build and deploy generative AI in different
ways in order to create more strategic benefits.
Introductory: Industry & Research Introduction
Sullivan Market Research
Enhanced Decision-
Making Capacity
Reduce Costs
8
www.leadleo.com
400-072-5588
Key findings
The best case selection process is vendor registration and participation, case input, case assignment, and
final case delivery and release. The case selection will combine industry cross big data with Sullivan's
innovative all-dimensional selection metrics to objectively and fairly present the best practices in the
generative AI industry.
Source: Sullivan
Best Practices Selection Methodology
Best Practices Selection Process
01 02 03 04
Manufacturer
Registration
Case
Input
Case
Assignments
Case
Release
•Manufacturer
registration: provide
basic information
such as industry and
contact information
•Case input: Sullivan
sends questionnaires to
vendors and conducts
in-depth market
research to understand
the case service
situation of each
vendor.
•Case assignment:
Combining the results of
the questionnaire with the
vendor's communication
information, assigns points
based on industry modules,
selection indicators
•Case Publication: Write the
best case selection content
according to the case
assignment, and maintain
transparent communication
with vendors to ensure the
accuracy of the case content.
Functionality and Adaptability
Demand Adaptability, Data
Supply Specialization, Scenario
Function Generalization
Performance and Innovation
Quality controllability of
generated content,
inference latency of
generated content,
compliance security of
generated content
Deployment and Support
Generating content reasoning costs
(operations), program deployment
costs (pre-deployment), time costs
for program implementation,
training and support
Experience and Satisfaction
Scenario value, experience and
customization, performance
and innovation
❑Methodology: Comprehensive assessment of case service capabilities of various modules in different
industries around four dimensions: functional value and applicability, technical performance and
innovation, implementation and support, and customer experience and satisfaction feedback.
Appraisal
Scope
Sullivan Market Research
0.3 Selection Process and Methodology
Introductory: Industry & Research Introduction
Sullivan blends traditional and innovative research methodologies, combines industry-crossing big data, and presents
the best cases in the generative AI industry objectively and impartially through diversified research methodologies
and innovative all-dimensional assessment metrics.
Chapter I
Scoring Dimension Analysis
❑Functionality and Adaptability Dimension
❑Performance and Innovation Dimension
❑Deployment and Support Dimension
❑Experience and Satisfaction Dimension
•Functional value and applicability dimensions measure whether the solution fits the needs of
practical applications, including the adaptability to the needs of the customer's enterprise,
the specialization of data supply, and the generalization of functions in different business
scenarios.
•The performance and innovation dimension focuses on measuring the core technical
capabilities and innovations included in the program at the level of generated content
benefits, specifically Controllability of the quality of generated content, generated content
inference latency and generated content compliance security.
•The deployment and support dimension evaluates the economics of the program based on
different customer usage scenarios and the vendor's ability to provide training and support
services in implementation, specifically including the cost of generating content reasoning,
program deployment costs, program implementation practice costs, and training and support
metrics.
•Experience and satisfaction Dimension intuitively judges the benefit value of the solution in
practical implementation through the perspective of enterprise customers, specifically
including Scenario value, experience and customization, performance and innovation
satisfaction.
10
www.leadleo.com
400-072-5588
Key findings
Functional value and applicability dimensions measure whether the solution fits the needs of practical
applications, including the adaptability to the needs of the customer's enterprise, the specialization of
data supply, and the generalization of functions in different business scenarios.
Scoring Analysis – Functionality and Adaptability
❑Definition of Functional Value and Applicability Dimension: Focus on the overall performance of the
program in realizing the expected functions, bringing benefits, resource management and multi-scenario
application to ensure that it has the ability of high efficiency, adaptability and strategic alignment, and
lay down the direction to be able to conform to the scenario requirements in the actual application.
Dimension
One
Level 1
Indicators Secondary Indicators Indicator Highlights
Demand
adaptabilit
y
Expected product goals, positive
returns, development and
utilization costs, long-term
strategic goals
Evaluate performance in terms of
goal attainment, benefit realization,
cost considerations and strategic
alignment to ensure that it adapts to
and addresses market needs
Data
supply
specializati
on
Training data input, data supply
program, data processing program,
new data processing utilization
Evaluate the ability to specialize
and optimize data provisioning to
ensure that the solution can
effectively support the needs of
specific scenarios
Scenario
Function
Generaliza
bility
Extensibility, rapid packaging of
new features, multi-language
support
Assess the ability of the program to
flexibly apply and provide efficient
solutions in different business
scenarios
Indicator Definitions
Success Criteria - Functionality and Adaptability
All three indicators are ahead of the
market average, highly in line with the
actual application scenarios demanded
The scores of the three indicators are all
at the average level, and overall basically
meets the requirements
Inability to address the clients needs of
the actual application scenarios
Source: Sullivan
Sullivan Market Research
Low Programmatic Capacity High Programmatic Capacity
AWS × CIMC
1.1 Functionality and Adaptability Dimension
•Demand Adaptability: AWS provides CIMC with a number of in-depth communication activities, digital employees built
through LLM on Amazon Bedrcok for the Group to comprehensively improve productivity, the on-line Financial Secretary
Assistant and the Maintenance Assistant built based on the Amazon Bedrock Knowledge Base to assist in intelligent Q&A for
enterprise employees.
•Data supply exclusivity: Proprietary data on CIMC's internal information such as maintenance documents are used for training
input, and the business data is entered into the data lake in a uniform manner.
•Scenario Functionality Generalization: The solution supports a multi-language environment, through the RAG knowledge
base, the Agent to create a digital employee to achieve multi-departmental connectivity within the group.
Chapter I: Scoring Dimension Analysis
E.g
11
www.leadleo.com
400-072-5588
Key findings
The performance and innovation dimension focuses on measuring the core technical capabilities and
innovations included in the program at the level of generated content benefits, specifically
Controllability of the quality of generated content, generated content inference latency and generated
content compliance security.
Scoring Analysis - Performance and Innovation
Technical Performance and
Innovation Dimension
Dimension
Two
❑Technical Performance and Innovation Dimension Definition: focuses on the core technical capabilities
and innovative performance of the program in terms of generation quality, inference speed, compliance
and security, which can demonstrate significant and effective value in real-world applications.
Level 1
indicators Secondary indicators Indicator highlights
Controllabil
ity of the
quality of
generated
content
Model algorithm selection and
evaluation, content generation
accuracy, generation quality
improvement program, generation
quality stability
Evaluate to ensure the high quality
and reliability of the generated content,
the controllable quality of the
generated content will determine the
shape of the product
Generate
Content
Inference
latency
Generating content reasoning
speed, reasoning response latency
reduction program
Evaluate the speed of reasoning to
ensure that content is generated;
reasoning latency will determine how
users use the product
Generate
content
Compliance
Security
Model Compliance Requirements
Compliance, Standing Automated
Monitoring, Model Security
Compliance Safeguards
Evaluate compliance measures to
ensure generated content, security will
determine user trust in the product
Source: Sullivan
Sullivan Market Research
•Generated Content Accuracy: Conch Robot has demonstrated high accuracy and user adoption in roles such as
office assistant and code assistant, with an accuracy rate of up to 88%, which significantly improves business
efficiency. In addition, Alibaba Cloud continues to improve the quality of content generation through the SFT fine-
tuning and the RAG knowledge base.
•Inference efficiency optimization: Alibaba Cloud designs optimized inference solutions for Hello, including but
not limited to inference framework performance optimization, API and terminal network optimization.
•Security Compliance: Enabling Minword Filtering with Alibaba Cloud Content Security API to achieve legal
compliance filing.
ALIBABA CLOUD × HELLO-INC
1.2 Performance and Innovation Dimension
Success Criteria - Performance and Innovation
All three indicators are ahead of the
market average, highly in line with the
actual application scenarios demanded
The scores of the three indicators are all
at the average level, and overall basically
meets the requirements
Inability to address the clients needs of
the actual application scenarios
Chapter I: Scoring Dimension Analysis
Indicator Definitions
Low Programmatic Capacity High Programmatic Capacity
E.g
12
www.leadleo.com
400-072-5588
Key findings
The deployment and support dimension evaluates the economics of the program based on different
customer usage scenarios and the vendor's ability to provide training and support services in
implementation, specifically including the cost of generating content reasoning, program deployment
costs, program implementation practice costs, and training and support metrics.
Scoring Analysis - Deployment and Support
❑Definition of landing implementation and service support dimension: The program's economy,
scalability, landing efficiency and continuous support are assessed to have high efficiency and reliability,
so as to meet the customer's needs in the actual application, and to enhance customer satisfaction and
the long-term value of the program.
Dimension
Three
Level 1
indicators Secondary Indicators Indicator Highlights
Generating
content
inference costs
(using
operations)
Cost per million tokens, reduced
inference cost scheme
Evaluate the cost of generating content
versus ongoing measures to reduce
inference costs. Costs will determine
application usage and iteration
frequency
Program
deployment
costs (pre-
deployment)
Infrastructure deployment cost
structure, cost optimization
center solution, capacity cap
expansion solution
Evaluate the initial acquisition cost of
the customer of interest and consider
expansion flexibility and cost
management capabilities
Time cost of
program
implementation
Generated content accuracy and
reliability, service stability and
sustainability, input and output
content compliance
Evaluation focuses on the time cycle to
achieve feature go-live based on the
vendor's understanding of the customer's
needs
Training and
Support
Unique innovative features,
innovative inputs and outputs,
iterative model update cycle
Evaluate the resources and services that
focus on the vendor's support of the
client's continued empowerment after
the implementation of the buildout
Source: Sullivan
Sullivan Market Research
ALIBABA CLOUD × Lenovo
1.3 Deployment and Support Dimension
Success Criteria - Deployment and Support
All three indicators are ahead of the
market average, highly in line with the
actual application scenarios demanded
The scores of the three indicators are all
at the average level, and overall basically
meets the requirements
Inability to address the clients needs of
the actual application scenarios
Chapter I: Scoring Dimension Analysis
E.g
Indicator Definitions
Low Programmatic Capacity High Programmatic Capacity
•Reasoning and deployment cost optimization: The solution meets Lenovo's performance and cost optimization
needs under different business scenarios by providing models of different sizes, which is very economical and
cost-effective. In addition, AliCloud provides Hundred Refinement API for Lenovo's capacity expansion, with no
capacity limit.
•Training and Support Services: At the solution service support level, Alibaba Cloud regularly arranges
communication activities for Lenovo, synchronizes model progress with the customer side, and cooperates to
optimize the internal model. In addition, it provides SLA 7x24-hour work order support and nail group support
with a minute response speed for cloud resources and LLMs respectively.
13
www.leadleo.com
400-072-5588
Key findings
Experience and satisfaction Dimension intuitively judges the benefit value of the solution in practical
implementation through the perspective of enterprise customers, specifically including Scenario value,
experience and customization, performance and innovation satisfaction.
Scoring Analysis - Experience and Satisfaction
Dimension
Four
❑Experience and Satisfaction Feedback Dimension Definition: Judge the overall value and customer
satisfaction of the solution in actual application from the user's perspective, providing a strong basis for
enterprises in selecting and optimizing GenAI solutions.
Level 1
Indicators Secondary indicators Indicator highlights
Scenario
value
Degree of customer business
understanding, degree of matching
customer needs, maturity and scale of
GenAI products, willingness to invest
in product solutions in the long term,
and degree of data governance
system improvement.
Evaluate whether the program
understands and meets the needs of the
case clients, delivers substantial business
enhancement, and is supported and
invested in over the long term.
Experience
and
customizatio
n
Interactive experience of the program,
integration of the program with
existing process systems,
effectiveness in meeting growth
needs
Evaluate whether the solution can provide
users with a user-friendly and intuitive
operating experience, integrate efficiently
with existing systems, and meet the
growth needs of the organization through
comprehensive support services.
Performance
and Creation
Satisfaction with the quality of
generated content, satisfaction with
the speed of generation, satisfaction
with the security compliance needs of
the organization, and the degree of
vendor innovation and uniqueness
met
Ensure that the program can provide an
efficient and reliable program in terms of
quality of generated content,
responsiveness, and security compliance
in real-world applications, taking into
account its sophistication and
innovativeness from the user's perspective
Source: Sullivan
SenseTime × Haitong Securities
1.4 Experience and Satisfaction Dimension
All three indicators are ahead of the
market average, highly in line with the
actual application scenarios demanded
The scores of the three indicators are all
at the average level, and overall basically
meets the requirements
Inability to address the clients needs of
the actual application scenarios
Sullivan Market Research Chapter I: Scoring Dimension Analysis
E.g
Indicator Definitions
Low Programmatic Capacity High Programmatic Capacity
Success Criteria - Experience and Satisfaction
•Satisfied with the value of the scenario: intelligent Q&A, code and other programs fully meet the expectations
of Haitong's needs, and significantly improve the efficiency of Haitong's internal work.
•Willingness to cooperate in the long term: the person in charge of Haitong Securities said that in the future, it
will combine with the full-stack AI capability of SenseNova to continue to jointly promote the securities
industry's business processes, interactive changes and digital intelligent business system reconstruction, and
explore the experience of LLMs in industry verticals.
Chapter II
Compendium of Best Practice for
Generative AI Industry
❑Collection and Analysis of Best Practices for Generative AI
❑State of Generative AI Core Technology Applications
❑Comprehensive Atlas of Best Practices for Generative AI
❑Challenges and Developments in Various Industries
❑Potential Application Risks of Generative AI in Various Industries
❑Best Practices in Various Industries
•At this stage, generative AI is first applied in data-intensive or high-tech industries such as
consumer, finance and healthcare. Due to the maturity of generative AI technology and the
primary need for enterprise digital transformation, the technology is now more focused on the
benefits of enterprise work efficiency improvement.
•The development of core technologies, exemplified by GANs, VAEs, Transformers, etc.., has
fueled various industries and multiple types of content generation applications, such as the
application of GANs in medical imaging and the application of VAEs in equipment inspection
in the manufacturing industry.
www.leadleo.com
400-072-5588 15
Key findings
At this stage, generative AI is first applied in data-intensive or high-tech industries such as consumer,
finance and healthcare. Due to the maturity of generative AI technology and the primary need for
enterprise digital transformation, the technology is now more focused on the benefits of enterprise
work efficiency improvement.
Distribution and Value of Generative AI by Industries
Sullivan Market Research
Product
Developm
ent and
Design
Efficiency
Promotion
in
Production
Workflows
Optimizati
on
Chain
Manageme
nt
Operations
& Decision
Making
Risk
Manageme
nt
Clients
Operation Service &
Sales
Consumer &
Retail
Gaming &
Entertainment
Manufacturing
Healthcare
Financial
ICT
Automotive
Public Service
Education
Enterprise
Application
Design & Production Operations & Management Customers and Services
Significant Demand Higher Demand Demand Exists
❑Generative AI gradually and the economic activities of various industries to produce a close
combination of high quality to empower the industrial efficiency of various industries, improving the
productivity of enterprises.
•Industry penetration: Generative AI models need to rely on larger data sets for training, so at this stage, generative AI is more
effective in data-intensive high-tech industries that are more accepting of new technologies.
•Main optimization benefits: It is found that generative AI is still mainly focused on improving the efficiency of the enterprise,
on the one hand, generative AI is still in a rapid development stage, and its ability to analyze and process large-scale datasets
and repetitive work is more mature; on the other hand, relative to the design of innovations, the work of improving the
efficiency of the enterprise is the primary task under the background of the digital transformation.
2.1 Collection and Analysis of Best Practices for
Generative AI
Chapter II: Compendium of Applied Practices
Source: Sullivan
www.leadleo.com
400-072-5588 16
Key findings
The development of core technologies, exemplified by GANs, VAEs, Transformers, etc.., has fueled
various industries and multiple types of content generation applications, such as the application of GANs
in medical imaging and the application of VAEs in equipment inspection in the manufacturing industry.
State of Generative AI Core Technology Applications
GANs
Core Technology Practical Application Cases
Image
Generation
•Commonly used in the media
and gaming industry to
generate realistic face images
•Enhancement of training
datasets with newly generated
data samples, commonly used
in medical imaging
Data
Enhancement Art
•Generate artworks by
training models to assist
artists and designers in
exploring new art forms
VAEs
Image
Generation
Data
Downgrading Anomaly
Detection
•VAEs mainly generate realistic
handwritten digital images,
and can enhance the diversity
of image generation by
sampling the potential space
•VAEs achieve effective data
compression and denoising
by compressing and
reconstructing the data
•Identify anomalous samples
that are significantly different
from the training data by
comparing reconstruction
errors,.
Autoregressive
Models
NLP Speech
Production Image
Generation
•Autoregressive models
generate text word by word
based on a given context,
mainly used in machine
translation, text generation,
etc..
•Generate high-quality speech
signals by modeling the
conditional probability of audio
samples, often used in speech
synthesis, editing and, etc..
•Generate high quality images
by capturing complex
dependencies in images,
commonly used in natural
image generation, etc..
Transformers
NLP Computer
Vision Cross-Modal
Generation
•Based on BERT and GPT
models, it significantly
improves the performance of
various tasks, which is
commonly used in dialog
systems and code generation.
•ViT processes image data
by dividing the image into
fixed-size chunks and
treating them as sequence
elements, commonly used
in image classification tasks
•Generate high-quality
content by combining
different modal data, often
used in advertising design,
art creation, etc..
❑Core technologies exemplified by GANs, VAEs, autoregressive models, and Transformers underpin
the application of generative AI to multiple types of content generation tasks, such as image, text,
and audio, in various industries.
Sullivan Market Research
•Generative Adversarial Networks GANs have certain advantages in image generation quality and detail processing, and are widely used in
game design and face image generation in film and television; VAEs excel in realistic data distribution, and are widely used in data
compression and denoising, as well as anomaly detection, which is more commonly used in the manufacturing industry; autoregressive
models are also prominent in the field of data generation as are VAEs, and autoregressions are more commonly used in text generation,
etc..; Transformers have made breakthroughs in natural language processing and multimodal generation, and are commonly used in
dialog systems. Autoregressive models and VAEs are also prominent in the field of data generation, with autoregression being more
commonly used in text generation, etc.. Transformers have made breakthroughs in natural language processing and multimodal
generation, and are commonly used in dialog systems.
2.2 State of Generative AI Core Technology Applications
Chapter II: Compendium of Applied Practices
Source:Baidu Cloud、Sullivan
www.leadleo.com
400-072-5588 17
Sullivan Market Research
Source: Sullivan
Financial
Education
2.3 Best Practices in Various Industries
Functionality
and Adaptability
Performance and
Innovation
Deployment
and
Support
Experience and
Satisfaction
Feedback
Chapter II: Compendium of Applied Practices
18
www.leadleo.com
400-072-5588
Challenges and Developments in
The Gaming and Entertainment
Industry and Best Practices
•Generative AI technology empowers the game entertainment industry to subvert the traditional
art production, game experience, operation, and marketing model, and leading technologies
such as resource super-scoring and automatic face pinching have been applied to the game
entertainment industry one after another to promote the technological innovation of the
industry.
•Generative AI empowers the various process stages of the game entertainment industry, in
which the product development stage, assisted by generative AI technology, accelerates the
enterprise creation process and significantly reduces the cost of art resource creation; in
addition, the risk of copyright issues for game entertainment can also be detected and analyzed
by generative AI.
•Generative AI is widely used in the development and creation of the game and entertainment
industry, but the creation of content, in addition to the general consideration of privacy and
data risks, a large number of short and fast game and entertainment content relying on the
relevant technology may lower the threshold of the industry, crowd out the industry's high-
quality resource space, and bring challenges to the industry.
www.leadleo.com
400-072-5588 19
Key findings
Generative AI technology empowers the game entertainment industry to subvert the traditional art
production, game experience, operation, and marketing model, and leading technologies such as
resource super-scoring and automatic face pinching have been applied to the game entertainment
industry one after another to promote the technological innovation of the industry.
Source: AI Lab, Sullivan
Industry's challenges and The New Form of “GenAI+Gaming and Entertainment"
Sullivan Market Research
Traditional
Mode
Generative AI Technology Goals
Arts
Game
Experience
Operation
Marketing
High
production
costs
Homogenization
Work in
disarray
Weak
Conversion
Save
Control
costs
•Speech-driven Mouth Animation Technology: Generative AI generates
corresponding expressions and mouth shapes based on speech,
dramatically reducing the cost of animating faces and mouth shapes.
•Stylized head model generation: Based on a small number of head
models that match the game style, batch generation of head models of
the same style, reduces the cost and shortens the production cycle.
New
Experience
Innovation
•Auto Face Pinch: Generative AI can find the optimal face pinch parameters
based on the images uploaded by the player-user and generate them in
the game with one click.
•Language Interaction: Users can customize relevant voice commands to
drive the game character, and the system recognizes and generates skills
based on the relevant voice.
Steady
•Game environment monitoring: Assist the platform in preventing
players from uploading illegal content and images by comparing and
monitoring text and voice content.
•Game Abduction Crackdown: Using abnormal behavior detection and
visual technology to monitor and crack down on game abductions and
cheating.
Accurate
Precision
marketing
•Face-swap generation marketing: some game entertainment platforms,
players can participate in activities, through the uploading of photos to
replace the poster face, to obtain exclusive their own game virtual image,
this type of activities greatly enhance the player's contact and
participation in the game.
❑Generative AI optimizes and improves the art cost, game experience, operation and marketing,
breaking through the major pain points of the traditional game entertainment model and bringing a
new breaking point for the industry.
•The traditional mode of the game entertainment industry involves the high cost of art resources production,
homogenization of game experience, operational complexity and marketing difficulties and other pain points,
generative AI through the use of content generation, language and voice processing and other technologies,
empowering the industry to reduce the cost of production, optimize the player's experience, and to promote
the technological innovation of the industry as a whole.
2.4.1 Challenges and Developments in The Gaming and
Entertainment Industry (1/2)
Reduce Costs
Increase Efficiency
Chapter II: Compendium of Applied Practices
www.leadleo.com
400-072-5588 20
Key findings
Generative AI empowers the various process stages of the game entertainment industry, in which the
product development stage, assisted by generative AI technology, accelerates the enterprise creation
process and significantly reduces the cost of art resource creation; in addition, the risk of copyright
issues for game entertainment can also be detected and analyzed by generative AI.
Source: Tencent Cloud, Sullivan
Sullivan Market Research
Generative AI Opportunity Points Graph
Scenarios
Value
Created
Explore &
Prototype
Testing
Phase Oneline &
Application
Development
Phase
Market Analysis:
Product Positioning:
Player Behavior Analysis
Market analysis
and forecasting
Competitive
Player Analysis
Identified core themes
Of the game
Product cost and cycle
time forecasting
Program planning support
Art, music prototyping
Product Creation:
Program Development:
Game Plot Design Aid
Game Scenario
Design Aid
Original
design assistance
3D Modeling
Generation
user interface design
Program code writing
assistant
Game function test: Upline control:
Compliance Testing:
Automated Test
Execution
Game Flaw Prediction
Analysis
Game NPC Interaction
Testing
Bug Fixes
Content Audit and
Compliance Testing
Data source
traceability
Algorithms and Security
Assessment
Early warning of
emergencies
On-line data
monitoring
Player Apps:
Game Balance Test
Player Behavior
Prediction
Intelligent Customer
Service
Dynamic Content
Updates
Player Community
Management
Content Recommender
System
Accelerating the Creative Process
Reduce the Cost of Producing Art
Improve Game Operation Efficiency
Optimize Players’ Experience
❑Generative AI empowers all stages of the development process in the game entertainment industry, in
which the product development stage accelerates the creation process of game entertainment
enterprises and significantly reduces the production cost of art resources under generative AI
technology.
1Exploration and prototype phase: Generative AI assists companies to better implement product positioning and program
planning through market and player data analysis.
2Development phase: Generative AI assists companies in this phase by dramatically shortening and reducing creation cycles
and art production costs through art design assistance.
3Testing phase: Generative AI can be used to automate testing to reduce compliance risk for the functionality and copyright risk
issues that are key considerations in the gaming and entertainment industry.
4On-line and application phase: based on the analysis and prediction of player behavior, generative AI optimizes the game
experience by enhancing the interactivity between NPCs and players, etc.
Copyright Compliance
Testing
Anti-Addiction Monitoring
for Minors
Information
collection
and
organization
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Source: Sullivan
Chapter II: Compendium of Applied Practices
2.4.1 Challenges and Developments in The Gaming and
Entertainment Industry (1/2)
Note: Some applications involve multiple technical functions, the corresponding colors just represent the main functional
modules.
Opportunity points
for generative AI
applications across
scenarios in the
Gaming and
Entertainment
industry
www.leadleo.com
400-072-5588 21
Key findings
Generative AI is widely used in the development and creation of the game and entertainment industry,
but the creation of content, in addition to the general consideration of privacy and data risks, a large
number of short and fast game and entertainment content relying on the relevant technology may lower
the threshold of the industry, crowd out the industry's high-quality resource space, and bring challenges
to the industry.
Source: Sullivan
Potential Application Risks
Content Creativity and
Compliance Risk Functional Areas
covered
Technology Dependence
and Autonomy Risk
User Privacy and Data
Security Risks
Market Adaptation and
Competitive Risk
Sullivan Market Research
Creative homogenization: AIGC may be overly reliant on pre-existing data and
models, leading to a lack of innovation and personalization in the content generated,
which reduces user experience and market competitiveness.
Compliance issues: As the content generated by AIGC may involve copyright, ethics
and legal standards, there is a need to ensure that the content generation process
strictly adheres to relevant regulations and industry standards.
Cultural sensitivity: In different cultural contexts, the content generated by AIGC may
mistakenly violate cultural taboos or be expressed inappropriately, causing
controversy and negative impact.
Game Plot
Generating
Over-reliance on technology: The gaming and entertainment industry may be over-
reliant on AIGC technology, leading to a loss of creators' autonomy and creativity
and affecting the sustainable development of the industry.
Iterative technology update: With the rapid development of AIGC technology, the
industry needs to continuously adapt to new technologies, which may lead to
challenges of insufficient resource investment and talent development.
Technical failures and security issues: The AIGC system may have technical failures or
security vulnerabilities that prevent it from detecting potential vulnerabilities and
issues in the game, affecting the quality of content generation and data security.
Game
Development
Artifact
User Interaction
Flat-Roofed
Building
Player Behavior Analysis and Misuse: While AIGC analyzes player behavior to
enhance the gaming experience, there is a possibility of misuse of the analysis
results, such as inappropriate marketing strategies.
Privacy invasion: the collection and use of player data may fail to adequately
protect the privacy rights of users, raising legal and ethical issues
Data security: generative AI training relies on a large amount of data including
player behavior data, and ensuring legal access and use of such data is a major
challenge
User Data
Analyze
Personalized
Recommender
System
Market adaptability issues: AIGC-generated content may not be in line with market
trends and user needs, leading to low market acceptance of products or services
Increased competition: With the popularization of AIGC technology, competition
within the industry is likely to intensify, and companies will need to continue
innovating in order to maintain a competitive edge
Market space crowding out: More game companies rely on AI technology to quickly
produce a large number of short and fast game content, crowding out the industry's
space for high-quality content, negatively affecting the market ecology
Market Analysis
and Projections
Automated Test
Game Plot
and Content
Generation
2.4.2 Potential Application Risks in The Gaming and
Entertainment Industry
Scene Characters
Creations
Content
Recommendation
Player Behavior
Analyze
Product
Positioning
Strategy
Chapter II: Compendium of Applied Practices
www.leadleo.com
400-072-5588 22
Sources: AWS, sullivan
Sullivan Market Research
A New Generation of AI-driven UGC 3D Interactive Content Platforms
Application Scenario: Gaming and Entertainment Industry + Generative AI-driven UGC 3D
Interactive Content Platform
Core Evaluation Keywords
Innovative Models of
Personalized Interaction
Low Cost,
Low Latency
Well-Established
Deployment and Support
Leading Comprehensive
Technical Performance
Capacity Analysis
Appraisal
Scope
•Solutions meet the existing practical needs: AWS assists BUDs
with efficient open source model deployment through
SageMaker's integration with Huggingface; and Claude3 Haiku
models on Amazon Bedrock enable low-cost and near-
instantaneous response for BUDs to build mimetic seamless AI
experiences that mimic human interactions.
•Fitting in with BUD's long-term strategic direction: The solution
provided by AWS is in line with BUD’s
long-term strategic goal of creating a disruptive
user experience in the gaming + social + meta
-universe through new technologies, and
partnering with BUD to create a new
frontier in gaming.
•Leading integrated technical performance: Amazon ensures
the high quality, stability and compliance of the output
content of the solutions provided to BUD with its leading
integrated capabilities in infrastructure behind the LLM, data
analytics and security compliance.
•Prompt engineering to optimize the innovative
product experience: Strongly structured, logical
prompt engineering assists BUDs in
achieving enhanced game character
interactions, innovations and optimizing
game backgrounds.
•Economy of implementation: Amazon
Bedrock is a fully hosted service that offers
many leading AI foundation models through a single
API, among them the Haiku model, which provides
a high-performance and affordable option for BUD.
•Completeness of Solution Implementation: The globalization
of AWS's resources assists BUD to better cover overseas users
in close proximity, and provides BUD with complete
documentation support, online and offline training, and
regular sharing sessions to respond to customers' needs for
support at any time.
•Customer Recognition of Existing
Solution: The solution provided by
AWS for BUD fully meets the customer’s
practical needs and expectations for cooperation.
Under this solution, BUD has achieved new
performance breakthroughs in terms of the number of
player interactions creations.
•Strong long-term cooperation and willingness to invest: BUD
and Amazon in addition to the existing optimization results,
game innovation level, both sides still have ongoing
innovation and experimental cooperation. BUD has a long-
term investment in the cooperation of strong will.
2.4.3 Best Practices in The Gaming and Entertainment
Industry- Next Generation AI-powered UGC 3D Interactive
Content Platform
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
SPECIAL DISCLAIMER: The foregoing specific AWS Generative AI related services are only available in Amazon Web Services overseas regions.
Specific information is subject to the official website of Amazon Web Services Overseas Region (aws.amazon.com).
Chapter II: Compendium of Applied Practices
www.leadleo.com
400-072-5588 23
Sullivan Market Research
Product Structure and Core Advantages
Client’s Demands
Effectiveness of Implementation
Solutions
Amazon
Transcribe
Convert Voice
to Text
Wisper on
SageMaker
Amazon Bedrock
Claude3 Haiku
Amazon Bedrock
Mixtral 7B
Amazon
Polly
Bark on
Amazon SageMaker
VPC
LLM Personalized
Voice
Vector
Database
Amazon
RDS Amazon
OpenSearch
Amazon
Lambda
Unity
Server
Game
Server
File
Embedding
Application Load
Balancer Amazon
S3
Amazon CloudFront Subscribers
Game
Players Game
Operation
123
4
2.1
1Convert Voice
to Text 2AI Chat, Games 3Generate
Voice 4Trigger 3D Character
Actions
2.1 Retrieve Chat
History
Solution Effectiveness
•Innovative experience: Enrich the platform content, so
that users have new gaming experience.
•Cost and time reduction: The BUD enterprise faces
time-consuming, labor-intensive, inflexible
expansion,etc.. For the optimization of prompt
engineering in different game scenarios, the exploration
time of model selection process needs to be
compressed.
•The Claude3 Haiku model on Amazon Bedrock: Cost-
effective and fast reasoning, and fast API calls using Bedrock
to meet customer needs
•Huggingface + Amazon SageMaker: More efficient open
source model deployment for BUD.
•Amazon Polly Generative Engine: Generates highly
personable speech while maintaining low latency.
•AI intelligences based on Amazon Bedrock Claude3 Haiku are more realistic at the level of contextualization and semantic
understanding, making users feel more connected.
•Games developed based on AI modules have greater randomness and interactive personalization.
•As of April 2024, the number of AI Buddy interactions reached 5 million, the number of AI Intelligences created exceeded 1.5
million, and the number of AI Intelligence interactions surpassed 50 million.
Chapter II: Compendium of Applied Practices
AI
Infra
Sources: AWS, sullivan
SPECIAL DISCLAIMER: The foregoing specific AWS Generative AI related services are only available in Amazon Web Services overseas regions.
Specific information is subject to the official website of Amazon Web Services Overseas Region (aws.amazon.com).
AWS has built a perfect content platform and ecosystem for BUD's overseas game business,
realizing an innovative AI-driven game model with high randomness and interactivity, and
assisting BUD to reach a new breakthrough in the number of overseas players.
AWS BUD
www.leadleo.com
400-072-5588 24
Sullivan Market Research
AI Native Virtual Social Ecology
Industry-leading Knowledge
Base Building Data Security Compliance Highly Customizable
Characters
Appraisal
Scope
•Demand Adaptation: In response to the demand for AI character
refinement in ZHUMENGDAO, SenseTime relies on the Sensechat-
character to provide users with a PUGC AI character production
process.
•IP character customization: Sensechat-character can quickly and
qualitatively complete customized IP characters, enabling industry-
leading character dialogues, personas, and plot promotion capabilities.
•Knowledge content specialization: negotiating the
availability of Sensechat-character’s ability to
build in-depth knowledge bases, it can accurately
respond to characterization, character relationships,
worldview, and event memory related content.
•Long Dialogue Memory: Sensechat-character combined with 32K
ultra-long context processing capability, with short and long term
memory fusion algorithm, Zhumengdao can realize accurate
memory of AI characters.
•Safety Compliance: SenseTime provides customers with technical
support for safety compliance related to
the landing application of LLMs. In
addition, SenseCore big device support
enterpris to deploy a proprietary cloud to
meet the needs of data security.
•Matrix of models adapted to different
application scenarios: SenseChat-Character
-Pro and SenseChat-Character, respectively, for
scenarios that require optimal results and scenario
that have high concurrency requirements.
•Stable and Reliable Reasoning Service: SenseTime provides
customers with standardized API products, enabling enterprises to
quickly build specialized LLM application with 99.5% availability.
•Deployment Cost: : SenseChat-Character has introduced different
versions of the model according to the needs to support the
optimization of the cost-effectiveness of customer enterprises.
•Satisfaction with the value of the
scenario:The relevant person in charge of
ZHUMENGDAO said that Sensechat-character
fully meets the expectations of enterprises at the level of AI
character setting fit and scene dialogue.
•Satisfaction with experience and customization: SenseTime and
ZHUMENGDAO worked together on R&D to give the popular IP
characters in ZHUMENGDAO more dynamic, emotional and other
abilities to optimize the user experience. In addition, the team of
ZHUMENGDAO said that during the period of cooperation : Sense
Time fully provided rapid response technical service support for
them.
Long Dialogue Memory
Source: SenseTime, Sullivan
Application Scenarios: Game entertainment industry + personalized character creation +
emotional accompaniment, film and television / animation / net article IP characters + star /
net celebrity / artist AI doppelganger + language role-playing and other anthropomorphic
dialogue scenarios
2.4.3 Best Practices in The Gaming and Entertainment
Industry - AI Native Virtual Social Ecology
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Chapter II: Compendium of Applied Practices
Core Evaluation Keywords
www.leadleo.com
400-072-5588 25
SenseTime ZHUMENGDAO
Source: SenseTime, Sullivan
Sullivan Market Research
Effectiveness of Implementation
•User Activity: The operation data of ZHUMENGDAO APP shows that the average daily use of users is 130 minutes, the number of chat rounds
per capita per day is as high as 135 rounds, the next day retention rate of users is more than 50%, and more than 100 head creators have
been formed.
•Accurate Persona Display: Relying on the ability to dynamically adjust the intimacy of SenseChat-character and set personalized
requirements, ZHUMENGDAO achieves a more accurate display of the character's persona as well as a more immersive interaction.
•Abundant AI characters: “ZHUMENGDAO”already has 150,000 AI characters, including well-known IP characters and original content.
Among them, popular IP characters, such as Xiao Yan and Xiao Meixian from “Battle Through the Heaven”, and top-stream virtual
characters such as Fan Xian, Emperor Qing and Uncle Wu Zhu from “Joy of Life ", are very popular.
•IP Character Refinement Demand: Zhumengdao is committed to
creating an AI native virtual social ecosystem for its users. In
order to provide users with more diversified interactive
experiences and imagination, ZHUMENGDAO has put forward a
more refined and higher-quality demand for the characters'
ability to fit their persona, dialogue ability and character
intimacy.
•SenseChat-character: SenseNova-based industry models include
industry-specific functions such as character creation customization,
knowledge base construction, multi-person chat, which assist
Zhumengdao in providing users with more dynamic and persona-
appropriate AI characters and more emotional interaction experiences,
creating a more diversified interactive experience and unlimited
imagination space for users.
Core Technical Advantages of the SenseChat-Character
IP Role
Customization
Services
Knowledge
Base
Construction
Personalized
Experience
Content Security
and
Service Guarantee
•Provide fast, high-quality IP
character customization
services according to the
specific scenarios and needs
of enterprises.
•Supporting persona command
attack defense mechanism,
the third party can not use the
enterprise exclusive
customized IP role, to avoid
potential risks.
•With knowledge base depth
construction capabilities, it
realizes accurate replies related to
character settings, character
relationships, worldviews.
•The system updates the
proprietary memory bank,
combining contextual processing
capabilities with short/long-term
memory fusion algorithms to
realize accurate memory for AI
characters.
•Supports multi-grade intimacy
settings for AI characters to create a
diverse dialogue experience and
meet personalized needs.
•It supports flexible adaptation and
fitting of user's emotions based on
chatting habits without departing
from the character settings,
providing a more realistic emotional
interaction experience.
•Provide safe and compliant
technical services for the
landing of enterprise LLM
applications.
•Based on the powerful AI
infrastructure of Sensecore, it
supports proprietary cloud
deployment, meets the needs
of enterprises in terms of data
security, and provides stable
and reliable services.
Product Structure and Core Advantages
Client’s Demands Solutions
Solution Effectiveness
Chapter II: Compendium of Applied Practices
Based on SenseChat-character, SenseTime and ZHUMENGDAO jointly realized the AI character
display with accurate persona, providing users with a more realistic and temperature-rich virtual
social experience, and the implementation of the program assisted Zhumengdao to achieve an
eye-catching performance in user activity and number of interactions.
26
www.leadleo.com
400-072-5588
Challenges and Developments in
The Manufacturing Industry and
Best Practices
•On the basis of the original application paradigm of the integration of manufacturing and
artificial intelligence, the generative and generalization capabilities of generative AI technology
support some of the complexity and innovative problems in the industrial manufacturing
industry to be solved, and will continue to expand the application space of artificial intelligence
in the manufacturing industry.
•Manufacturing as the core of the industrial manufacturing industry has been the existence of
environmental protection and safety of the two major industry challenges, generative AI in the
environmental analysis and assessment as well as the production of risk monitoring level to
empower enterprises to optimize the point; in addition, generative AI in the logistics and
production management links, enhanced to assist the original system to further enhance the
process of various aspects of management efficiency.
•Industrial manufacturing segments for the operation of the accuracy and real-time requirements
of high, generative AI generated content deviation may bring serious risks to the safety of
workers and equipment; in addition, the industrial data structure is diverse and complex, the
enterprise in the pre-training and deployment of generative AI resource cost is extremely high
and the input-output ratio is not yet clear.
www.leadleo.com
400-072-5588 27
Key findings
On the basis of the original application paradigm of the integration of manufacturing and artificial
intelligence, the generative and generalization capabilities of generative AI technology support some of
the complexity and innovative problems in the industrial manufacturing industry to be solved, and will
continue to expand the application space of artificial intelligence in the manufacturing industry.
Source: Tencent Cloud, MIIT, Sullivan
Sullivan Market Research
"Generative AI +
Manufacturing"
Algorithms and
Training Tools
Generic Support
Technology
Industrial Knowledge
and Experience
Deep
Learning
Machine
Learning
Knowledge
Graph
Knowledge
Engineering
Reinforcement
Learning
...
Edge
Computing Cloud
Computing
Data Security Data Platform
Visualization ...
Expert
Experience Design
Verification
Engineering
Experience Simulation
Industrial
Network ...
❑Generative AI meets specific needs in different parts of industrial manufacturing, and the ability to
generate and generalize new knowledge and insights supports the industrial manufacturing industry
in solving a broader range of more complex challenges.
•Before the integration of generative AI into the manufacturing industry, AI functions such as quality inspection and equipment
maintenance prediction have been deeply applied to the manufacturing industry, and a mature application paradigm has been formed,
including algorithms and training tools, general support technologies, and industrial knowledge and experience adapted to manufacturing
scenarios.
•With the evolution of generative AI macromodels, specific needs have been added to the application paradigm of industrial manufacturing.
For example, at the level of algorithms and training tools, architectures such as Transformer have laid the foundation for generative AI to
enter the manufacturing industry; at the level of general support technologies, vector databases and other critical digital infrastructures; at
the level of industrial knowledge and experience, generative AI‘s requirements for high-quality data such as text, images, and documents
are increasing. Under the original application paradigm, the integration of generative AI technology solves more complex and extensive
manufacturing difficulties, will continue to expand the application space of AI in the manufacturing industry.
•The traditional industrial manufacturing industry faces challenges such as production costs, environmental governance, and risks to the
safety of workers and equipment. With the integration of artificial intelligence into all aspects of the manufacturing industry, artificial
intelligence empowers the overall process of forecasting, production, management, and decision-making, helping industrial enterprises to
reduce costs and increase efficiency. According to data from the Ministry of Industry and Information Technology, with the empowerment of
artificial intelligence in the manufacturing industry, the industry's research and development cycle has been shortened by about 20.7%,
production efficiency has been improved by about 34.8%, the rate of defective products has been reduced by about 27.4%, and carbon
emissions have been reduced by about 21.2%.
Industry's challenges and The New Form of “GenAI+Manufacturing"
2.5.1 Challenges and Developments in The
Manufacturing Industry (1/2)
Chapter II: Compendium of Applied Practices
www.leadleo.com
400-072-5588 28
Sullivan Market Research
Generative AI Opportunity Points Graph
Value
Created
R&D Design
and
Planning
Products &
Services
Market Analysis:
Product Design:
Market Analysis and
Forecasting
Product Design
Assistance
Equipment
Management：
Intelligent Scheduling
Equipment Fault Diagnosis
Prediction
Production Quality
Control
Industrial Production
Code Assistance
Supply Chain
Management: Product Marketing:
Staff Management
Training:
Product Service:
Accelerate Product Innovation Improve production quality and
operational reliability
R&D and design
software assistance
Scientific Creation of Eco-
Friendly Materials
Production Process
Optimization：
Intelligent Assembly
Robot
Production process
optimization
Supply Chain Data
Analytics
Supply Chain
Optimization
recommendations
Operations
Management:
Knowledge Quiz
Assistant
Operations Management
Decision Support
Production Training
Material Generation
Production Energy
Consumption Management
Marketing Content
Generation
Personalized Marketing
Content Analysis
Intelligent Customer
Service Assistant
Intelligent Product
Showcase
Sales Trend Analysis and
Forecasting
Supply Network
Design
Logistics Information
Analysis
Inventory Status
Monitoring
Optimize warehousing logistics and scheduling
management to improve production efficiency
Enhancing industrial safety and environmental standards to ensure safe and sustainable production processes
❑Generative AI effectively mitigates environmental and safety issues in the manufacturing industry. At the
logistics management and production level, generative AI combines with the industry's original system to
empower enterprises to enhance the efficiency of scheduling and management in all aspects.
1R&D design and planning: AIGC combines the results of market analysis and introduces image generation capabilities into industrial design
scenarios such as CFM design and channel customization.
2Manufacturing: AIGC is used in the core manufacturing process for shop floor equipment management to improve safety and productivity
and optimize the overall workflow.
3Operation Management: Realize intelligent analysis of management data through natural language interaction and other ways, and empower
CRM and other management software to improve the quality of supply chain management.
4Product service: AIGC empowers products to achieve intelligent interaction and personalization and differentiation in product marketing and
after-sales service.
Key findings
Manufacturing as the core of the industrial manufacturing industry has been the existence of
environmental protection and safety of the two major industry challenges, generative AI in the
environmental analysis and assessment as well as the production of risk monitoring level to empower
enterprises to optimize the point; in addition, generative AI in the logistics and production management
links, enhanced to assist the original system to further enhance the process of various aspects of
management efficiency.
Production Process
Monitoring
2.5.1 Challenges and Developments in The
Manufacturing Industry (2/2)
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Information
collection
and
organization
Manufacturing Business
Management
Source: Sullivan
Chapter II: Compendium of Applied Practices
Note: Some applications involve multiple technical functions, the corresponding colors just represent the main functional
modules.
Scenarios
Opportunity points
for generative AI
applications across
scenarios in the
Manufacturing
industry
www.leadleo.com
400-072-5588 29
Key findings
Industrial manufacturing segments for the operation of the accuracy and real-time requirements of high,
generative AI generated content deviation may bring serious risks to the safety of workers and
equipment; in addition, the industrial data structure is diverse and complex, the enterprise in the pre-
training and deployment of generative AI resource cost is extremely high and the input-output ratio is
not yet clear.
Source: Sullivan
Potential Application Risks
Technology Dependency Risk Functional Areas
covered
Attribution Risk
Supply Chain Risks
Application Challenges
Sullivan Market Research
Robustness: Generative AI models have the potential to fail to recognize details in
equipment inspection, resulting in outputs that cannot be correlated to real-world
environments, leading to inaccurate or risky behavioral decisions, including unhealthy
impacts on future maintenance and management planning for equipment assets.
New Material
Developmental
Design
Product
Development
Design Aid
Supply discrimination: When generative AI is used in the evaluation of suppliers in
the supply chain management chain, there is a risk that bias in the data or models
may lead to unfair screening, and companies should improve the transparency of this
through, for example, fair contract terms.
Training data quality: the industrial manufacturing field covers a wide range of data
structure is diverse and complex, generative AI application data before spending a lot
of time and resources for data clarity, preprocessing and calibration.
Application real-time: Many production scenarios in the manufacturing industry have
high real-time requirements, requiring generative AI application models to respond in
milliseconds and microseconds.
Production
Process
Make Superior
Intelligent
Assembly
Robot
Supply Chains
Data Analysis
Reliability: For the industrial manufacturing industry, the accuracy of the output
content is particularly important, inaccuracy of the model output results may bring
hidden dangers and serious industrial risk results for worker safety, normal operation
of equipment, etc..
Equipment
Failure
Sensing
Production
Process
Monitors
Equipment
Maintenance
Attribution of environmental responsibility: The new materials and
products designed by generative AI have the potential to negatively impact
the environment, and it is not possible to clearly delineate the subject of
responsibility for long-term negative impacts on the environment.
Liability attribution for accidents: lack of clear guidelines and procedures for
penalizing liability for shop floor accidents caused by incorrect information
and advice from generative AI.
Supply Chains
Data Analysis
Supplier Credit Risk: Problems with data or models when generative AI evaluates
supplier credit can lead to the selection of unreliable suppliers and affect supply chain
stability.
Supply chain disruptions: the application of generative AI in supply chain
management can lead to supply chain disruptions, affecting production and delivery,
if predictions or decisions are wrong.
Supply Chains
Optimization
Recommendations
Supply
Network
Devise
Input-output ratio: the training of generative AI models and the cost of private
deployment is expensive, and the current industrial application of generative AI is still
in the primary stage, for which the input-output ratio is not clear.
2.5.2 Potential Application Risks in The Manufacturing
Industry
Production
Process
Monitors
Chapter II: Compendium of Applied Practices
www.leadleo.com
400-072-5588 30
Sullivan Market Research
CIMC Generative AI Platform
Application Scenarios: Manufacturing + generative AI empowers functional applications to improve
human efficiency + scenario-based applications (financial secretary assistant, maintenance assistant, etc..)
Highly Convergent and
Integrated Innovation in
Cooperation Data Security and
Compliance One-Stop Shop for
Efficiency Gains
•Demand Adaptation: AWS built digital employees through the
big language model on Amazon Bedrcok for CIMC to
comprehensively improve productivity. Also the on-line Financial
Secretary Assistant and the Maintenance Assistant built based on
the Amazon Bedrock Knowledge Base to assist in intelligent Q&A
for enterprise employees.
•Data supply exclusivity: Proprietary data on CIMC's internal
information are used for training input, and the business data
entered into the data lake in a uniform manner.
•Scenario Functionality Generalization: The solution is
very useful in supporting multiple language
environment, through the RAG knowledge base,
creating digital workforce to realize multi-sectoral
connectivity.
•Controlled quality of generated content: AWS improves the
quality of generated content through data quality proofreading,
dataset categorization, and hybrid search, and ensures the stability
of the generated quality through content traceability, user scoring,
and RAG revision.
•Low inference latency: The solution reduces the inference
response latency of the generated content by means of contextual
filtering, frequently asked questions, and high-frequency content
RAG.
•Compliance and Security: All data in Amazon
Bedrock is encrypted in transit for CIMIC to
ensure that user data privacy and security.
•Upfront Deployment and Implementation Cost
Optimization: Amazon offers Bedrock managed
Services combined with API Gateway and Lambda,
providing CIMC with capacity limit expansion. Also,
Provide API unified release to call related capabilities to
integrate into CIMC's internal system.
•Training and Support: AWS provides CIMC with comprehensive
functional operation documentation guidelines, relevant regular
training activities and round-the-clock support services.
•Scenario Value Satisfaction: Amazon
and CIMC based on the "Joint Laboratory
Mechanism", jointly constructed the a mature programme
that combines the actual needs of enterprises and achieves a
comprehensive improvement in internal productivity.
•Experience and customization satisfaction: CIMC generative AI
platform efficiently integrates the enterprise's existing system
platforms, integrates all kinds of processes, and basically reaches
barrier-free use.
•Satisfactory performance and innovation: The solution is based
on the 6+3 technical, which simultaneously meets the security
compliance requirements as well as the guarantee of the
performance of LLM capabilities.
2.5.3 Best Practices in The Manufacturing Industry
- CIMC Generative AI Platform
Chapter II: Compendium of Applied Practices
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Core Evaluation Keywords
Sources: AWS, sullivan
SPECIAL DISCLAIMER: The foregoing specific AWS Generative AI related services are only available in Amazon Web Services overseas regions.
Specific information is subject to the official website of Amazon Web Services Overseas Region (aws.amazon.com).
Appraisal
Scope
31
www.leadleo.com
400-072-5588
Sullivan Market Research
Amazon
S3
Data
Manage-
ment
Amazon Redshift
CIMC-DPM
Generative AI Applications
Solution Effectiveness
•Innovative Technology Pursuit: CIMC attaches great
importance to technological innovation and hopes to utilize
generative AI to drive business innovation.
•Generative AI application landing is difficult: LLM has general
knowledge without specialization, "hundred model mixing",
related technology update iteration is too fast and other
difficult problems, become generative AI in the CIMC within
the landing application of the hurdle.
•"6+3" framework design: Amazon Cloud and CIMC jointly designed the
"6+3" generative AI application framework for enterprises, integrating
and applying multiple models, intelligently triaging and selecting the
appropriate model according to business scenarios and requirements,
and achieving a good balance between user experience, performance
and cost while ensuring enterprise safety and compliance. The user
experience, performance and cost are well-balanced, while ensuring
enterprise security and compliance.
•Currently based on the generative AI ability platform to create onboarding of various functions of the digital staff has more than 10, all kinds
of enhancement effect > 100.
•The Financial Secretary's Assistant is online, providing a fast response within 30 seconds, 24/7, with an answer adoption rate of 99.6%.
•With the help of the maintenance assistant, the downtime for "critical failures" has been reduced by 20% and the familiarization time for
new employees has been reduced from one year to six months.
•Promoting the popularity of CIMC's generative AI applications, 98% of employees think they can improve efficiency for their jobs.
Amazon
Glue
Amazon
S3
Amazon OpenSearch
Vector Database
Generative AI Service
Subnet
Amazon
SageMaker
Amazon Bedrock
Amazon
SageMaker
Endpoint
Generative AI API Subnet RAG
Amazon API
Gateway Amazon
lambda
LLM Routing
Efficiency
Financial
Manu
-facturing
Compliance -
Check
Bidding
Application Load
Balancer Digital
Workforce Maintenance
Assistant
Financial
Secretary's
Assistant
Chapter II: Compendium of Applied Practices
Effectiveness of Implementation
Client’s Demands Solutions
Product Structure and Core Advantages
Sources: AWS, sullivan
SPECIAL DISCLAIMER: The foregoing specific AWS Generative AI related services are only available in Amazon Web Services overseas regions.
Specific information is subject to the official website of Amazon Web Services Overseas Region (aws.amazon.com).
Combined with CIMC's internal practical needs, AWS has co-designed and constructed an
enterprise-grade generative AI capability base to support three types of applications, namely
digital employee, maintenance assistant and finance assistant, to comprehensively improve
enterprise productivity, operational efficiency and maintenance efficiency.
AWS CIMC
www.leadleo.com
400-072-5588 32
Sources: Alnnovation, Sullivan
Sullivan Market Research
Miracle Automation Digital Display Platform Based on LLMs
Application Scenario: Manufacturing Industry+ Intelligent Q&A and Data Query System
Professional Data
Information Analysis Innovation of Cooperation Labor Cost Reduction and
Efficiency Generate Accurate Content
Appraisal
Scope
•Requirement Adaptation: Improved the efficiency of employee
knowledge querying from an average of 10 minutes to less than 1
minute; saved 80% of data analysts' working time.
•Data supply specialization: Self-service uploading classification
can be realized for customer document knowledge, vector
database is used to realize document cut management, and all
kinds of databases are linked by address to realize direct natural
language dialogue query.
•Generalization of scenario functions: The programme is targeted
at different business sectors and can be quickly applied by
uploading the sector's industry knowledge base at the
user's own discretion.
•Generated content quality can be controlled: the accuracy rate of
data content generation is more than 90%, and the accuracy rate
of knowledge quiz generation is 95%.
•Low inference latency: for the text generation class, the first
token starts outputting in 1.5s; the code generation function has
an inference speed of 1s, including data retrieval as well as
answer generation.
•Compliance and safety: in algorithm design, training data
selection, model generation and optimization, and to take
measures to prevent racial, ethnic and religious
discrimination on the basis of race, creed, etc..;
the generated content is in line with the
socialist core values.
•Optimization of solution deployment costs:
Use load balancing techniques such as DNS polling,
hardware load balancers or software load balancers to
reasonably distribute user requests, continuously monitor
application and server performance indicators, and optimize and
adjust resources based on monitoring results.
•Training and support: A innovation provides customers with on-
site training demonstrations, answers to frequently asked
questions and 24/7 service support.
•Experience and customization: The
interface design is simple and beautiful,
the operation process is smooth and the style is unified, so it
can be used quickly and efficiently; the services provided by
the vendor are very comprehensive and efficient, supporting
the long-term development of the business.
•Performance and Innovation: The Generative AI solution
demonstrates the vendor's high level of innovation and uniqueness
in technology and functionality, providing industry-leading and
innovative features and solutions that significantly differentiate it
from other products in the marketplace and bring a clear competitive
advantage.
2.5.3 Best Practices in The Manufacturing Industry
- Miracle Automation Digital Display Platform Based on
LLMs
Chapter II: Compendium of Applied Practices
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Core Evaluation Keywords
www.leadleo.com
400-072-5588 33
Sullivan Market Research
•Difficulty in accessing and analyzing data: Data analysis
needs need to be provided by professionals, and companies
do not have access to real-time data analysis and reporting.
•High cost of knowledge acquisition: The number of
knowledge documents within the enterprise is large, and the
cost of knowledge acquisition is relatively high.
•Difficulty in real-time querying: Enterprise personnel cannot
query internal knowledge and data in real-time and need to
access internal systems for knowledge acquisition.
•ChatDoc knowledge Q&A application, with excellent knowledge
extraction, positioning, summarization and reasoning capabilities, can
help enterprises quickly achieve intelligent Q&A, summarization and
answer traceability of internal knowledge.
•ChatBI data analysis applications, support for a variety of formats of
data sources, support for the user's full process can be intervened,
editable, confirmable, data analysis results are reliable and credible;
Text-To-SQL, Text-To-Chart ability to reduce the threshold of data
analysis, enhance the efficiency of data analysis.
•Generated content quality can be controlled: the accuracy rate of data content generation is more than 90%,and the accuracy rate of
knowledge quiz generation is 95%.
•Improved efficiency: Employee knowledge query efficiency, from an average of 10 minutes of query time to less than 1 minute.
•Reduce operation difficulty: Simple dialog can be real-time query of all production data, without professional data engineers can quickly
build data Kanban, saving 80% of the data analysts' work time.
•Enhance competitiveness: More convenient equipment control and more flexible data insights reduce reliance on manual operations
and help enhance market competitiveness.
Intelligent production system for automated production line in pilot plant
Data source
introduction MySQL CSV/Excel PDF Doc TXT
Knowledge
Learning
Common
knowledge
Statement
knowledge digital
dictionaries sector
knowledge
Knowledge Tags
synonyms ES
Recall Vector
recall
information retrieval
fusion
algorithm
knowledge management
Knowledge
upload knowledge
consult
knowledge
examine
knowledge
compiler
data processing
digital
withdraw
original text
slice
concern
generating
indexing
construct
model analysis AInnoGC Industrial Large Model
Q&A analysis
intelligent
Q&A Chart
Generation Report
Generation
Issue
linkage historical
record Chart
Export
Answer
Generation Answer
Tracebility
Detail
Inquiry
text
summary Q&A
History Evaluation of
answers
ChatBI Data Analytics Application ChatDoc Knowledge Quiz System Identity and Access
Account Management
User Authentication
User Authentication
Rights Management
Monitoring and
Maintenance
Resource Management
Log Management
Event Management
Monitor Alarms
Enterprise MES system and knowledge base
Product Structure and Core Advantages
Solution Effectiveness
Sources: Alnnovation, Sullivan
Chapter II: Compendium of Applied Practices
Configuration Management
Effectiveness of Implementation
Client’s Demands Solutions
In response to the practical needs of Miracle Automation's internal process optimization,
Alnnovation provided the company with an intelligent interaction and digital display platform
based on LLM, which optimized the process of the company's workshop as well as the efficiency
of the company's office in an all-round way.
Alnnovation Miracle Automation
34
www.leadleo.com
400-072-5588
Challenges and Developments in
The Healthcare Industry and
Best Practices
•Generative AI's ability to generate data and information assists pharmaceutical companies to
break through the customary difficulties of new drug development and listing, and personalized
medical advice helps the industry to shift from passive medical treatment to a precise and active
medical treatment model; in addition, generative AI solves the privacy problem of doctor-patient
information data through synthetic data.
•Generative AI brings a series of value benefits to the healthcare industry, such as pharmaceutical
R&D innovation, management efficiency optimization, medical research and service support,
customized marketing, etc.., of which pharmaceutical innovation and medical service
optimization are the key areas of the current generative AI layout.
•While boosting the transformation and upgrading of the healthcare industry, generative AI also
brings challenges such as medical content bias, liability and accident attribution, patient privacy
and security protection, fairness and transparency, and whether the new model incorporating
generative AI is effective or not still needs to be confirmed by a large number of clinical trials and
multi-party verification.
www.leadleo.com
400-072-5588 35
Source: NVIDIA, Sullivan
❑The healthcare industry faces limitations such as high regulation, confidentiality of doctor-patient
information sharing, and upgraded demand for patient service experience. Generative AI provides
the industry with more accurate medical advice and anonymized doctor-patient information data
support through the generation of new data and information.
Key findings
Generative AI's ability to generate data and information assists pharmaceutical companies to break
through the customary difficulties of new drug development and listing, and personalized medical advice
helps the industry to shift from passive medical treatment to a precise and active medical treatment
model; in addition, generative AI solves the privacy problem of doctor-patient information data through
synthetic data.
Sullivan Market Research
"Generative Capabilities"
to Empower the
Front Office
Foreground
Middle and Back
Office
Marketing
& Sales
Medical Care
Research and
Development
Production
and Quality
Information
Data
Management
Legal Affairs
Compliance
Human
Resource
"Smart data"
Empowers the Middle
and Back Office
Traditional Front Office
Challenges:
•Tough regulatory environment for drug
products
•Risks of exposing private information by
asking for a consultation
•Difficulty in pricing products
•Inaccurate marketing positioning
•Patient Service Experience Requirement
Enhancement
•Quality control and safety management risks
Traditional Middle and Back
Office Challenges:
•Medical product innovation is
limited
•Difficulties in information
distribution and management
integration
•Difficulty in training medical
personnel
•Medical Services and Devices Legal
Compliance Changes
Generative AI Benefits
Medical Research
Generative AI assists medical
professionals in understanding
disease mechanisms through
biological process simulation
Video and Image
Processing
Video and image enhancement
and processing assist
physicians in making more
accurate disease diagnoses
Data
Anonymization
Generative AI can safeguard
data privacy by anonymizing
data in cases where
confidentiality is required
Industry's challenges and The New Form of “GenAI+Healthcare"
•Traditional challenges in the industry: Healthcare is one of the highly regulated industries, with high market access thresholds for drugs
and devices, so companies face challenges in new drug development, launching and pricing; in addition, the accuracy of medical
diagnosis and the privacy of doctor-patient data sharing are also traditional difficulties in the industry.
•The new form of “Generative AI + Healthcare”: Compared with traditional AI, generative AI generates more creative and accurate
content based on data processing and analysis capabilities, helping pharmaceutical companies accelerate the efficiency of new drug
R&D in the process of market launch and boosting the process of market access; in addition, the generative capabilities also focus on
patient service and marketing levels. Besides, the generative capability also focuses on patient service and marketing, providing patients
with personalized marketing plans and consultation suggestions through accurate and targeted diagnostic suggestions, helping to
promote the healthcare field from the traditional passive medical treatment to active medical treatment mode; finally, for the use and
sharing of sensitive doctor-patient information and data, generative AI can achieve data sharing by synthesizing data under the premise
of safeguarding data privacy.
Drugs Market
Access Study of
Medicine
Strategy &
Operations
2.6.1 Challenges and Developments in The Healthcare
Industry (1/2)
Chapter II: Compendium of Applied Practices
www.leadleo.com
400-072-5588 36
Sullivan Market Research
Pharmaceutical Innovation
R&D
Medical
Services and
Sales
Post-Diagnosis
Supports
Business
Operational
Management
2.6.1 Challenges and Developments in The Healthcare
Industry (2/2)
New Drug Development:
Potential Drug Target
Mining
Production and Supply Chain
Management Diagnostic services Health Management:
Precision Marketing:
Personalized follow-up
visits:
Accelerating new drug innovation to market
Improve production quality and operational reliability
Generation of new
chemical structures
Medical Image
Synthesis Personalized Health
Management Plan
Health data monitoring
and analysis
Automatic
generation of case
records
More proactive and targeted treatment services
More accurate medical information co-construction, intelligent consultation and precise marketing to optimize the doctor-
patient experience
1Pharmaceutical Innovation R&D: GenAI can assist pharmaceutical companies in exploring potential drug targets and new chemical
structures, shortening the drug development cycle and marketing process.
2Enterprise Operations Management: GenAI helps companies strengthen control at the drug product quality and supply chain levels through
contextual understanding and data analytics.
3Medical services and sales: generative AI improves the efficiency of medical services and the doctor-patient experience through the rapid
collection of medical data, and provides customized marketing plans based on patient information.
4Post-diagnosis support services: Through real-time interaction, applications such as digital human health education and publicity are integrated
into post-diagnosis support to provide better health management for patients.
Forecasting the
feasibility of new drugs
Toxicology, predictive
risk assessment
New Drugs on The
Market:
Clinical Trial Data Analysis
Forecasting
Optimized Design of
Clinical Trials
Product Pricing
Decision Support
Registration Information
and Audit Analysis
Laboratory reagent
equipment
monitoring
Sample screening and
quality analysis
Quality training
content generation
Quality inspection
report review
Talent training
Training and exam
content generation
Digital Human Training
Screening of
Professionals
Supply chain optimization
recommendations
Disease treatment
and decision support
Supportive Diagnosis
and Prediction
Treatment program
content generation
Rare Disease Analysis and
Abnormal Alerts
Patient Information
Update Record
Precise
classification of
customer groups
Automated Answers
to Doctor's Questions
Vital signs data
monitoring
Patient education
content generation
Digital Human Patient
Education Campaign
Follow-up report
content generation
Online follow-up
communication
Patient medication
task reminders
Optimization and coordination of medical resource allocation
❑Based on the risk and high accuracy requirements, generative AI directly used in surgical operations
use case is still relatively small, new drug development assistance and medical service optimization for
generative AI currently focus on the layout of the field.
Key findings
Generative AI brings a series of value benefits to the healthcare industry, such as pharmaceutical R&D
innovation, management efficiency optimization, medical research and service support, customized
marketing, etc.., of which pharmaceutical innovation and medical service optimization are the key areas
of the current generative AI layout.
Generative AI Opportunity Points Graph
Note: The application involves multiple technical functions, and the corresponding colors represent the main functional technical
modules.
Source: Baidu Cloud, Sullivan
Chapter II: Compendium of Applied Practices
Opportunity points
for generative AI
applications across
scenarios in the
Healthcare
industry
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Information
collection
and
organization
Value
Created
Scenarios
www.leadleo.com
400-072-5588 37
Key findings
While boosting the transformation and upgrading of the healthcare industry, generative AI also brings
challenges such as medical content bias, liability and accident attribution, patient privacy and security
protection, fairness and transparency, and whether the new model incorporating generative AI is
effective or not still needs to be confirmed by a large number of clinical trials and multi-party verification.
Source: Sullivan
Potential Application Risks
Functional Areas
covered
Sullivan Market Research
Complementary
Diagnosis
and Projections
Precise Customer
Base
Categorization
Disease
Treatment and
Decision Support
Treatment Plan
Content
Generation
Clinical Trial Data
Analysis
Forecasting
Medical Record
Automatic
Generation
Patient
Information
Update A Record
2.6.2 Potential Application Risks in The Healthcare Industry
Risks to The Reliability of
Generated Content
Weak Coding and Erroneous Output: Generative AI suffers from logic gaps and
comprehension biases, and may be overcoded or undercoded, bringing in erroneous
information that can lead to unreliable output results.
Risk of Bias: The potential for bias in the data used in the training of generative AI
models, leading to the generation of biased algorithms for the unequal treatment of
patients from different demographic groups, e.g., studies have shown inaccuracies
in generative AI diagnostic results for patients with darker skin tones.
Attribution of Responsibility
and Accountability Risks
Attribution of Liability for Technological Misdiagnosis: Generative AI may
be misdiagnosed in medical services due to the limitations of the training
data as well as bias, and the mechanism for assuming liability for the
resulting medical risk damages is not clear.
Doctor-Patient Trust Crisis: Biased information generated by generative AI
may cause doctors to give wrong decisions or fail to communicate
effectively with patients, which in turn causes doctor-patient suspicion and
triggers consumer trust risks. Privacy and Security
Risks
Illegal Access and Output of Information: All patient information read by
generative AI is protected by law, and wrongful access to data and output of
information will bring issues of privacy and security and legal disputes.
Privacy Breach: Patient medical privacy information will be directly exposed to the
underlying model during the pre-authorization and coding assessment process,
which may pose the risk of illegal third-party access and information leakage.
Security Vulnerability: There are security vulnerabilities in the storage and
transmission of medical data in generative AI.
Transparency and
Validation Challenges
Application Validation: Whether the "generative AI + healthcare" model is
effective and better than the traditional healthcare industry needs to be
confirmed by a large number of clinical trials and multiple validations.
Transparency of Algorithms: In the case of medical advice, for example, a
lack of transparency of the relevant algorithms may infringe on the patient's
right to information and choice.
Disease
Treatment and
Decision
Support
Treatment Plan
Content
Generation
Health Data
Monitoring and
Analysis
Treatment Plan
Content
Generation
Clinical Trial Data
Analysis
Forecasting
Complementary
Diagnosis
and Projections
Disease Treatment and
Decision Support
Chapter II: Compendium of Applied Practices
www.leadleo.com
400-072-5588 38
Source: Inspur, Sullivan
Sullivan Market Research
Intelligent Generation and Sharing Platform for Electronic Medical Records
Application Scenario: Medical and Healthcare Industry + Intelligent Generation and Sharing
Platform for Electronic Medical Records
Core evaluation keywords
Intelligent Medical Record
Generation
High Quality of Text
Generation Compliance Assurance Optimization of The Doctor-
Patient Process
3.6.3 Best Practices in The Healthcare Industry
- Intelligent Generation and Sharing Platform for
Electronic Medical Records
•Demand Adaptation: Inspur is highly adapted to customer
demands, automatically generating medical record
information and significantly reducing the workload of
doctors.
•Data supply specialization: Specialized knowledge
datasets, section-specific datasets, and customer-private
high-quality labeled data were used.
•Scenario Function Generalization:
Customers can utilize new proprietary
data for enhanced training based on
Model Ops and Agent Store to
implement new functional applications.
•Controlled quality of generated content: 90% accuracy of
generated content, 80% user adoption rate, high quality
control.
•Low inference latency: The inference speed (response
latency) of the generated content is 1 second, showing a
faster inference speed.
•Compliance and safety: Meet the compliance
requirements of the National Internet
Information Office algorithm filing and LLM
on-line filing, and the normalization
of automatic monitoring flaws safeguard
the content.
•Generating content inference costs: In the
direction of model lightweighting on the
optimization, which effectively reduces the
inference cost.
•Solution Deployment Costs: The localized deployment
uses GPU servers and achieves the expansion of the
capacity limit through load balancing, showing good
scalability and flexibility.
•Time cost of program implementation: The time cycle
from project creation to functional validation POC was 30
days, and the time cycle from POC to formal go-live was 2
months, showing a faster implementation speed.
Business satisfaction with the program:
C u s t o m e r s a t i s f a c t i o n Inspur 's
understanding of business and process
requirements, as well as the solution's
scenario suitability, was highly evaluated.
•Maturity and scale of the program: The GenAI program is
currently being tested and validated on a small scale, and
has not yet been fully applied to formal production, but
shows a certain degree of maturity.
•Willingness to invest in the long term: The First Affiliated
Hospital of SDFMU has a strong willingness to invest in
generative AI technology in the long term, regards it as a
strategic technology, and plans to continue to increase
investment and resources.
Chapter II: Compendium of Applied Practices
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Appraisal
Scope
www.leadleo.com
400-072-5588 39
Source: Inspur, Sullivan
Sullivan Market Research
•Time-consuming collection of information from doctors and
patients: Doctors need to collect information on patients' signs
and condition development during their hospitalization, it is
time-consuming to organize all the records and other
paperwork.
•Optimization of hospital staff's office process needs: for a large
number of repetitive paperwork collection and collation work,
hospitals need to optimize the repetitive workflow with new
technology to improve the efficiency.
•Intelligent Generation and Sharing Platform for Electronic Medical
Records: Through the operation of cell phone APP in the process of
consultation and the acquisition of voice technology for doctor-
patient dialogues, it realizes the rapid generation of instruments and
supports the synchronization of the doctor's modification and
confirmation to the electronic medical record system.
Product Structure and Core Advantages
Solution Effectiveness
•The work efficiency is significantly improved, greatly reducing the workload of doctors and helping them to reduce their workload input by
70%.
•It guarantees the security and compliance of the platform and meets the encryption and auditing requirements for medical data.
•The simplified design of the user interface and the ease of use of the system allows physicians to get started quickly and easily.
The solution built on the Inspur Hai Ruo medical model
Distributed Arithmetic Platform
Resource
pooling
Resource
dispatch Heterogeneous
computing
Operations
management Computational
framework
Large model of the Hajoki base
Data
management
Model
development Cluster
Management Distributed
training Inference
service
Medical Atomic Capabilities
Medical entity
extraction
Medical Semantic
Recognition Clinical
reasoning
Medical Text
Generation
Medical Atomic Capabilities
Medical entity
extraction Medical Semantic
Recognition Clinical reasoning Medical Text
Generation
•Hai Ruo writes medical
records.
•Hai Ruo does research.
Complementary medicine
Intelligent triage (medicine)
•Medication
supervision
•Blood control
Grade Review Assistant
Performance Appraisal
Assistant
Model
Ops
Full life cycle
managerial
Expert support
LLMin position
LLM Review
LLM Training
Data
Development
+
LLM
Security
Models can be
corroborated
Healthcare
Commission Model
Hospital Model
Model of Health
Insurance
Data not out
of domain
Healthcare
Commission
Patients
Medical insurance
Core
Advantages
Data-driven
strategic
decision
Cloud resource
enjoy together
Safety
Compliance
safeguards
Technical
expertise
be in favor of
Chapter II: Compendium of Applied Practices
Effectiveness of Implementation
Client’s Demands Solutions
Inspur cooperated with the First Affiliated Hospital of Shandong First Medical University to realize
the intelligent generation of electronic medical records through generative AI technology, which
comprehensively improved the efficiency of medical document processing for the hospital,
ensured data security and compliance, optimized doctors' workflow, and enhanced the patient
service experience.
Inspur The First Affiliated Hospital of Shandong First
Medical University
www.leadleo.com
400-072-5588 40
Source: Tencent Cloud, Sullivan
Sullivan Market Research
Ruijin Hospital Medical Model
Application Scenario: Healthcare Industry + Medical Intelligence Q&A + Follow-up
Recommendation Generation
Core evaluation keywords
Medical Intelligence Quiz High Quality of Text
Generation
Follow-Up Proposal
Generation
Optimization of The Doctor-
Patient Process
2.6.3 Best Practices in The Healthcare Industry
- Ruijin Hospital Medical Model
•Demand Adaptation: Tencent Cloud, in response to the practical
needs of Ruijin Hospital, helped Ruijin Hospital build the Ruijin
Hospital Medical Model, in which medical intelligent Q&A and
follow-up recommendations were generated to meet the needs
of the hospital's internal medical staff to improve their work
efficiency.
•Data supply exclusivity: Tencent Cloud's medical model is based
on a hybrid grand model adding
more than 2.85 million medical entities, 12.5
million medical relationships, and more than
98% of the medical knowledge of the
knowledge breakthroughs and Medical
Literature.
•Compliance and Security: Based on Tencent Cloud’s Hunyuan
model base, it strictly follows domestic and international laws
and regulations, such as the European Union's General Data
Protection Regulation and China's Cybersecurity Law, to ensure
that the security and privacy of users' data are fully protected.
•Controllable quality of generated content: the program is based
on Tencent’s Hunyuan model, which is regularly evaluated, and
the model is evaluated according to the evaluation feedback
results. Each performance is continuously optimized and
upgraded.
•Scenario Value Satisfaction:
The solution provided by Tencent Cloud for Ruijin
Hospital effectively and efficiently alleviated a
large amount of repetitive writing work for healthcare
professionals and significantly improved the efficiency of
healthcare professionals within the hospital.
•Customers' willingness to invest in the long term: Tencent Cloud
and Ruijin Hospital's goals and are highly compatible, and Ruijin
Hospital has indicated that it will continue to carry out long-term
strategic cooperation to promote the development of a leading
medical model.
•Perfect service technical support:
Tencent Cloud has rich experience in fine
tuning industry LLM with regent
requirements consulting services, solution design,
landing the full process of program implementation,
delivery of project management projects.
Chapter II: Compendium of Applied Practices
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Appraisal
Scope
www.leadleo.com
400-072-5588 41
Source: Tencent Cloud, Sullivan
Sullivan Market Research
•Increased efficiency of medical staff within the hospital: After using Follow-up Suggestion Generation, Ruijin Hospital's Follow-up
Suggestion Generation reached a clinically usable level, and Ruijin Hospital internally saved more than 50% of the physicians' writing time.
Product Structure and Core Advantages
Solution Effectiveness
Core Advantages Based on Tencent Cloud Medical LLM
Click here to add textClick here to add textClick here to add
textClick here to add textClick here to add text
Click here to add textClick here to add textClick here to add
textClick here to add textClick here to add text
Base model
Medical
Literature
Fine-tuning of
medical missions
send back
information
intensify
Pre-training the model to master the underlying
medical knowledge
•Train generative large language base models at multiple
scales to meet the needs of different business scenarios
•The language model modeling process includes data on a
variety of topics, genres, languages, and channels, laying
the foundation for adapting downstream tasks
Ongoing pre-training in the medical field to further
the model's medical knowledge
•Large amount of high quality medical literature
•Large-scale medical knowledge graph content covering
98% of disease knowledge in ICDs
Fine-tuning of the base model based on data from multiple
medical tasks
•Based on the pre-trained model, introduce downstream business
for multi-task fine-tuning to improve the model landing effect
Feedback-enhanced incentives based on feedback to make model
responses more professional and doctor-like
•Training reward models by scoring and ranking model results by physicians
•Based on reward models, using reinforcement learning algorithms, optimization
models
•Complexity of the Q&A process: A large number of
patients asking basic questions about their condition
generates a lot of repetitive Q&A work for healthcare
professionals.
•Time-consuming Diagnostic Recommendation Writing:
Doctors spend a lot of time writing recommendations and
gathering information about the patient during the process
of generating a diagnostic recommendation for the patient.
•Medical Intelligent Q&A: Based on the common medical conditions in
medicine, intelligent Q&A, systematic medical knowledge mapping
and deep semantic analysis, it meet the patient's precise and
personalized information acquisition requirements.
•Follow-up suggestion generation:It can assist doctors in generating
customized follow-up plans and follow-up suggestion content
generation for in-depth patient management.
Chapter II: Compendium of Applied Practices
Effectiveness of Implementation
Client’s Demands Solutions
Tencent Cloud cooperated with Ruijin Hospital to build a medical model, in which intelligent Q&A
and follow-up recommendations generate service solutions that help Ruijin Hospital save more
than 50% of the internal doctors' writing time, dramatically improving the efficiency of the
hospital's internal work, and optimizing the patient-physician process experience.
Tencent Cloud Ruijin Hospital
42
www.leadleo.com
400-072-5588
Challenges and Developments in
The Financial Industry and Best
Practices
•The financial industry's unique high-intensity data attributes and the dynamic changes in the
economic environment bring significant challenges in data analysis and prediction and data
governance, and the difficulty of data analysis and prediction also limits the rationalization of
customer service recommendations. The innovative generation capability of generative AI helps
the financial industry to realize efficient and low-cost function expansion and data information
processing.
•Generative AI enhances data connectivity in the financial industry, assisting institutions and
analysts in shortening the length of time for massive data processing and analysis, improving the
efficiency and accuracy of data processing decisions, and empowering institutions to identify risks
such as fraud and abnormal data.
•Behind the high efficiency of data processing which enabled by generative AI, there are still
potential risks such as sensitive data leakage, model bias and overfitting, etc., which bring about
trading decision-making errors, so financial institutions and related departments need to establish a
perfect data security management system and carry out secondary validation and verification of
important information.
43
www.leadleo.com
400-072-5588
Key findings
The financial industry's unique high-intensity data attributes and the dynamic changes in the economic
environment bring significant challenges in data analysis and prediction and data governance, and the
difficulty of data analysis and prediction also limits the rationalization of customer service
recommendations. The innovative generation capability of generative AI helps the financial industry to
realize efficient and low-cost function expansion and data information processing.
Industry's challenges and The New Form of “GenAI+Finnace"
Source: Du Xiaoman, Tencent Cloud, Sullivan
Sullivan Market Research
2.7.1 Challenges and Developments in The Financial
Industry (1/2)
Traditional Financial
Industry Pain Points
Changes in the regulatory
environment
•Changing regulatory policies
•Increased difficulty in cross-
border regulation
Structural changes in
society
•Labor market shifts, high cost of
talent in finance
•Increase in the proportion of low
net worth clients
Risk management
challenges
•Risk of human error
•Risk of system failure
•credit risk
•Customer transaction data
privacy and security
Structural changes in
financial institutions
•Derivatization of financial
structures and
fragmentation of data
across institutions
Generative AI New
Business Matching
"Generative AI+Finance"
functionality
Financial
Fraud
Detection
Intelligent
Investment
Research
Precision
marketing
Changes in Customer
Behavior
•Diversification and personalization of
customer needs
•Shift in customer product acquisition
channels
Credit Risk
Control and
Security
Compliance
❑Generative AI assists the financial industry to break through the challenges of the changing regulatory
environment and complex risk management in the traditional model through innovative data models,
strategies, and simulations of complex markets.
Computer vision
•Processing of image information collected by
front-end hardware based on visual
perception and content extraction and
analysis
•Related applications: identity verification,
fraud detection, intelligent security can
Intelligent Voice and
Conversational AI
•Based on NLP and other technologies,
human-computer voice interaction core,
processing, generating human voice
•Related applications: voice intelligent
customer service, voice identity
verification
Knowledge map
•Building an association network based
on financial big data
•Related applications: risk management,
compliance checking, credit approval,
investment analysis
Natural language
processing (NLP)
•With techniques such as NLU at the core,
computers understand process and
generate natural language
•Related applications: intelligent
customer service, market research
•Industry Traditional Challenges: The financial industry's unique high-intensity data attributes and the
dynamic changes in the economic environment make it challenging to analyze data, risk and data
management.
•New form of "Generative AI+Finance": Based on knowledge graph, natural language processing and other
technologies, generative AI creates new data models, strategies and personalized marketing content, and
realizes risk monitoring and management by simulating complex market environments.
Chapter II: Compendium of Applied Practices
44
www.leadleo.com
400-072-5588
Key findings
Generative AI enhances data connectivity in the financial industry, assisting institutions and analysts in
shortening the length of time for massive data processing and analysis, improving the efficiency and
accuracy of data processing decisions, and empowering institutions to identify risks such as fraud and
abnormal data.
Sources: DuXiaoman, Sullivan
Generative AI Opportunity Points Graph
❑Generative AI empowers financial institutions, enhances connectivity between data, shortens the time to
analyze massive amounts of complex data, and improves the efficiency of resource allocation, while
strengthening transaction risk control within the market and institutions.
1Risk control: Generative AI can predict and identify potential risks more quickly and comprehensively, allowing financial institutions to make
the right decisions quickly to reduce losses. In banks, it can be used for financial fraud detection and comparison of loan documents, etc.; in
brokerage organizations, it can be used to identify and review materials.
2Data analysis: generative AI can quickly preprocess data, data mining and predictive analysis, etc., providing strong support for decision-
making in financial institutions. It can be used for transaction data analysis in brokerage organizations; health data and claim prediction can
be analyzed in the insurance field.
3Product management: generative AI can enhance product innovation and can provide reasonable product pricing suggestions and more
accurate product promotion strategies. In asset management it can provide personalized investment advice as well as financial product
innovation.
Banks Share
Broker Insurance Asset
Management
Note: Some applications involve multiple technical functions, and the corresponding colors represent the main functional technical modules.
Risk Control:
Staff Empowerment:
Service Advice
Risk control:
Data Analysis:
Smart Trading:
Risk Control:
Product Design and
Optimization:
Data Analysis:
Risk Control:
Asset Management:
Product Management:
Enhance risk control capacity and help financial institutions to take timely and effective measures to reduce risk and loss
Improve analysts' real-time and accuracy in analyzing trading data
Optimize the financial services solutions offered to clients
Smart Office
Financial Fraud Detection
Credit risk alerts
Pricing of
financial products
Bank Digital Staff
Personalized
Customer Service
Inquiry Service
Financial Services
Advice
Recommendations for
credit support programs
Risk warning
Material Identification
and Audit
Research Report
Generation
Trading data analysis
Investment Research
Assistant
Intelligent Investment
Services
Trading Robot Assistant
Risk warning
Recommended
Insurance Products
Personalized
Insurance Advice
Insurance Product
Design
Insurance Consultant
Claims Process
Optimization
Claims Prediction
Analysis
Health Data Analytics
Risk Warning
Credit scoring and
default prediction
Investment
Compliance Check
Asset allocation
optimization
Asset Pricing Modeling
Assistance
Intelligent Post-Investment
Management Report
Financial Product Innovation
Personalized
Investment Advice
Sullivan Market Research
2.7.1 Challenges and Developments in The Financial Industry
(2/2)
Chapter II: Compendium of Applied Practices
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Information
collection
and
organization
Value
Created
Scenarios
Opportunity points
for generative AI
applications across
scenarios in the
Financial
industry
45
www.leadleo.com
400-072-5588
Key findings
Behind the high efficiency of data processing which enabled by generative AI, there are still potential
risks such as sensitive data leakage, model bias and overfitting, etc., which bring about trading decision-
making errors, so financial institutions and related departments need to establish a perfect data security
management system and carry out secondary validation and verification of important information.
Source: Sullivan
Potential Application Risks
Risks to The Reliability of
Generated Content Functional areas
covered
Market Decision Bias: Financial services companies may suffer from inaccurate or incorrect
market judgments generated by generative AI that can affect the analytical decisions.
Trading Data
Analysis
Over-Reliance on Generated Data: Finance belongs to a highly data-intensive industry, and
there are certain limitations on the scope of data that generative AI can synthesize, and over-
reliance on synthesized financial data and may trigger invalid output results.
Liability Attribution Risks
Intellectual Property Risks: The demand for large amounts of training data carries legal risks
associated with intellectual property or copyright infringement, and organizations deploying
models need to assess the IP compliance of their training sets.
Privacy and Data Security Risks
Operational and Technical Risks
Risk Event Disclosure: If a generative AI discloses a risk event that leads to a financial institution
making an incorrect customer service decision, the rights and responsibilities of the relevant
stakeholders should be documented and delineated.
Risk of Sensitive Data Leakage: When training models and testing datasets, enterprises should
try to safeguard data security and conceal the relevant data, otherwise the model might be
attacked for leaking relevant confidential data due to malfunction or orientation.
Smart
Investment
Data Access Rights: In risk management and other aspects, data governance and analysis need to
be shared among various departments, and for some sensitive transaction data, inappropriate
rights restriction management will lead to data misuse and leakage.
Risk of Customer Information Leakage: Generative AI needs to generate personalized financial
solutions through a large amount of customers' personal financial information, in which there
is a risk of customer privacy data leakage. Enterprises should take measures, including model
access rights management,to protect relevant customer information
Market Manipulation Risk: Generative AI may be utilized to generate misleading information
that can influence market sentiment and prices, ultimately creating the risk of market
manipulation.
Misuse of Technology: Generative AI can be used for unethical or illegal purposes, such as
creating false financial reports or engaging in financial fraud, and needs to be jointly regulated
by the relevant authorities and agencies.
Sullivan Market Research
2.6.2 Potential Application Risks in The Financial Industry
Deviation in The Simulation of Insurance Scenarios: In the process of claims prediction, the
accuracy of insurance scenario simulation is particularly important, as deviations in the
generated content can lead to incorrect claims payment, resulting in two-way losses for both
the company and the customer.
Research Report
Generation
Claims
Prediction
Analysis
Risk
warning
Material
Identification
and Audit
Trading Data
Analysis
Personalized
Investment
Suggestions
User Health
Data Analysis
Trading Data
Analysis
Trading Robot
Helper
Risk
warning
Chapter II: Compendium of Applied Practices
46
www.leadleo.com
400-072-5588
Source: SenseTime, Sullivan
Chapter II. Compendium of applied practices
Sullivan Market Research
LLM of Finance
Application Scenarios: Financial Industry + Intelligent Q&A + Compliance and Risk Control +
Code Assistance + Office Assistant + Intelligent Research Reports
Core evaluation keywords
Financial Multimodal Full
Stack LLM Data Information Security
Protection
Customized Corporate
Services
•Data Specificity: Based on its own leading LLM technology, combined
with the real application scenarios within Haitong Securities,
SenseTime has constructed a complete chain of thinking in line with
the financial industry, in which the intelligent R&D assistant launched
based on SenseTime's Raccoon provides developers with intelligent
supplementation and dialog Q&A services through the Haitong's rich
internal data accumulation and in-depth understanding of the
customer's internal business logic.
•Long-term development strategy fit: Haitong Securities has always
insist on Technology+Data+Scenario"-drivenas
force to comprehensively strengthen business
development, and the solution provided by business
intelligence fits the company's internal development
strategy.
•Controlled generation quality: Take intelligent code assistant as an
example, the solution continuously improves the accuracy of
generated content based on error detection, performance
optimization suggestions and user feedback.
•Intelligent Q&A Assistant: Based on the powerful language
understanding and interaction capabilities of the SenseNova Model,
superimposed on the financial industry-related knowledge and
search engines, it can accurately understand the questions and
efficiently give replies so as to
improve productivity for internal employees.
•Intelligent R&D assistant: Assist developers in code
programming, provide code completion and dialog Q&A
services, reduce the threshold of development
technology, improve development efficiency and
enhance the quality of software delivery.
•Upfront Deployment and Reasoning Costs: Based
on SenseNova Model - the first "cloud, end,
edge“full-stack Big modeling product matrix that
meets the different scale scenarios of the customer’s
requirements.
•Covering all aspects of securities trading in the pre-, mid- and post-
trading phases: Through the leading LLM technology, combined with
the real application scenarios of Haitong Securities, SenseTime has
constructed a complete chain of thinking that conforms to the
financial industry, focusing on the landing of intelligent Q&A,
intelligent research and development, and intelligent research and
reporting and other business segments.
•Scenario value satisfaction: Intelligent Q&A,
intelligent code and other programs are fully in line
with the expectations of the Haitong demand, which
significantly improved efficiency within Haitong.
•Satisfactory feedback: Through deep cooperation with SenseTime,
Haitong Securities combine the full-stack AI capabilities to jointly
promote the business process, interaction changes and digital business
system reconstruction in the securities industry, and explore the
experience for the landing of LLMs in industry verticals.
•Willingness to cooperate in the long term: The person in charge of
Haitong Securities said that in the future, Haitong Securities will take the
multi-modal large model of Business Intelligence as the base capability,
and will deeply explore the potential of a wider range of applications in
finance, securities and other related industries, and will help Haitong
Securities to improve its operational efficiency, risk control capability and
customer experience.
2.7.3 Best Practices in The Financial Industry
- LLM of Finance
Improving The Quality and
Efficiency of Business Processes
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Appraisal
Scope
47
www.leadleo.com
400-072-5588
Source: SenseTime, Sullivan
Sullivan Market Research
Product Structure and Core Advantages– SenseTime Raccoon
Solution Effectiveness
•Business Efficiency Improvement: Based on Raccoon's intelligent R&D assistant, Haitong saves a lot of repetitive development work, unifies
the development behavior, code specification and comment style, and finally Haitong Securities achieves 40% development efficiency
improvement.
•Customer Experience Enhancement: In addition, Business Intelligence has enhanced digital human and other capabilities for Haitong,
helping more customers realize more innovative, timely and accurate service experiences such as intelligent customer service and intelligent
marketing.
•R&D Efficiency Challenges: Haitong Securities has a huge business scale and
faces a series of business process challenges in daily software development,
such as high percentage of repetitive work, uneven level of developers, and
unchanged internal information query, etc. The software development
projects are getting more and more complicated, and the overall
development efficiency of the development team is gradually becoming a
bottleneck for business development.
•Data information protection requirements: The financial industry is a data-
dense industry and data information involves a large amount of transaction
data and other highly confidential business information, so there are high
requirements for the security of data information within the enterprise.
•Haitong Securities and Raccoon have jointly launched "eHay
Yandao", an intelligent R&D assistant, which realizes a number of
technological breakthroughs in the software R&D process, including
code dialogue, completion, translation, refactoring, unit test
generation, etc., significantly improving the efficiency and quality of
R&D.
Raccoon Full Link Assist Advantage
Software
Development
Life Cycle
Sub-
process
Mould
abilities
Demand
Analysis
Architectural
Design Coding Software
Testing
Deployment
Online
System
Maintenance
Demand
Research
Needs
assessment
Requirements
specification
design
prototyping
•Many rounds of
dialogue
•logical inference
•Long text
comprehension
•content creation
•Expansion and
Optimization
•emotional
analysis
•...
technical
architecture
business
process
user
Module Functions
Coding
Realization
unit test
code review
version
management
system
development
integration
test
system testing
Assessment of
results
environmental
preparation
System
Installation
Acceptance
testing
User training
Technical
Support
bug fix
system upgrade
performance
optimization
•Vincennes
diagram
•many rounds of
dialogue
•logical inference
•...
•Annotation
Generation Code
•code completion
•code refactoring
•code translation
•Code generation
comments
•...
•Annotation
Generation Code
•code completion
•Generating Test
Cases
•many rounds of
dialogue
•...
•Annotation
Generation Code
•code completion
•Generating Test
Cases
•Long text
comprehension
•...
•Code refactoring
•many rounds of
dialogue
•emotional
analysis
•...
Effectiveness of Implementation
Client’s Demands Solutions
Chapter II: Compendium of Applied Practices
In response to the industry and enterprise pain points such as complicated work content and
highly intensive transaction data, SenseTime and Haitong have jointly built a new multi-modal,
full-stack ecosystem for the industry based on the SenseNova model, in which Raccoon assists
Haitong Securities in realizing the efficiency of development and meeting the needs of internal
security protection and code data confidentiality.
SenseTime Haitong Securities
48
www.leadleo.com
400-072-5588
2.7.3 Best Practices in The Financial Industry
- Wealth Management Assistant
Source:ZhipuAI, Sullivan
Sullivan Market Research
Wealth Management Assistant
Core evaluation keywords
Precise Recognition of
User Intent One-Stop Financial
Services
Data Security
Compliance
Financial Data
Specialization
•Requirements adaptability: Zahngle Fortune Path, a collaboration
between Zhipuai and Huatai Securities, builds a new generation of
wealth management assistants, accurately recognizes customers'
needs, solves the shortcomings of the traditional system in terms
of intent recognition and multiple rounds of interactions, and
effectively improves customers' experience of using the project
and the ability of one-stop wealth management services.
•Dedicated data supply: the project uses a large amount of
specialized financial data (e.g. financial books, information,
regulations, announcements of listed companies), formed
through incremental training and instruction fine-tuning
dedicated Huatai Financial Model 1.0, which
makes high accuracy in handling finance-related tasks
and professionalism, demonstrating significant
financial field advantage.
•Controllability of the quality of generated content: GLM model
ensures the quality and professionalism of the generated content
by overlaying specialized financial data and instruction set fine-
tuning, and has a strong capability of generating content in the
financial field.
•Compliance and Security of Generated Content: As a homegrown
self-developed model, Smart Spectrum GLM supports local private
deployment, ensuring data security and
Content compliance greatly enhances the credibility of the
program.
•Generating content reasoning costs (using
operations): project may involve high computational
costs in the reasoning process,especially when dealing
with multi-round interactions and complex tasks.
•Solution Deployment Costs (Upfront Deployment): Upfront
deployment costs can be high due to the large amount of data
processing and customization involved.
•Time cost of program implementation: The implementation of
the project requires some time cost, especially in the data
preparation, model training and deployment phases.
•Training and Support: The full range of support and technical
training provided by Zhipuai ensures that the Huatai Securities
team is able to fully utilize the system's features.
•Scenario value satisfaction: customer
feedback shows that Huatai Financial Model 1.0
makes the end-user experience better
by improving intent recognition and multi-round interaction
capabilities, solving the problem that traditional technology
is unable to accurately recognize the intention and carry out
multiple rounds of interaction with customers, possesseing
the ability to strengthen the interaction experience with
users and one-stop service.
•Performance and Innovation Satisfaction: The project excels in
technical performance and innovativeness, especially in the ability
of semantic understanding of multiple rounds and multiple intents,
as well as the accurate and effective summarization of documents,
which is leading the domestic and international modeling
capabilities.
Application Scenario: Financial Services Industry + Intelligent Customer Service + Wealth
Management Assistant
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Chapter II: Compendium of Applied Practices
Appraisal
Scope
49
www.leadleo.com
400-072-5588
Source: ZhipuAI, Sullivan
Sullivan Market Research
Product Structure and Core Advantages
Solution Effectiveness
•Under this program, the Huatai Financial Model 1.0 improves 10% to 20% in terms of effectiveness compared to the generic model.
•Intent Recognition and Multi-Round Interaction Challenges: Isolated
product forms, insufficient generalization of intent recognition, and
lack of multi-round conversation comprehension still exist, resulting in
a poor customer experience. Systems that can understand multi-round
conversations and provide accurate responses are urgently needed to
enhance the customer experience.
•Data Security and Compliance Needs: Data security and compliance
are critical when dealing with sensitive financial data. Clients need to
ensure data security and system compliance, especially in privately
deployed environments.
•Intelligent Customer Service and Wealth Management Assistant: Through
the application of Zhipuai‘s GLM series models, Huatai Securities has
constructed a new generation of intelligent customer service and wealth
management assistants with powerful intent recognition and multi-round
interaction capabilities, which are able to accurately understand customer
needs and provide professional financial advice and services.
•Local Private Deployment and Data Security: The project supports local
private deployment, which ensures the safe storage and isolation of sensitive
data and provides a highly secure solution.
Based on the Smart Spectrum AIGLM family of models
Zhipuai GLM Series Models
All Tools
GPTS
Picture
comprehension
dialogues
coding
Vincennes
diagram
Search
Enhancement
GLM - 4
CogVLM
ChatGLM
CodeGeeX
CogView
WebGLM
Model training
Financial
Books
Financial
Information
Financial
regulations
Financial
Bulletin
...
Model fine-tuning
invest and
study
first aid
...
40-50G
financial
expertise
overlay
A collection of
tens of
thousands of
high-quality
multi-scenario
commands
App "Zhanle Fortune Path"
Chapter II: Compendium of Applied Practices
Effectiveness of Implementation
Client’s Demands Solutions
Based on the GLM series, Huatai Securities and Zhipuai cooperated to develop wealth
management assistant, which aims to respond to the complex and changing business needs of
the financial industry, provide users with more accurate and efficient wealth management
services, and help financial institutions reduce costs and increase efficiency.
ZhipuAI Huatai Securities
50
www.leadleo.com
400-072-5588
Challenges and Developments in
The ICT Industry and Best
Practices
•Generative AI assists the information and communications industry in breaking through the complexity
of traditional manual operation and maintenance management difficulties, driving the industry to
intelligent network mode transformation, enabling the multi-dimensional resources for joint optimization
scheduling. In addition to promoting the information and communications industry more accurately fit
other industry sectors, expanding and updating of the business model.
•Generative AI is mainly applied to the operation and security management in the communications
industry at this stage, which enhances the automation of network operation by transforming big data
resources into efficient network planning and fault diagnosis capabilities through intelligent algorithms.
•The deployment and optimization of communication networks involves the ecological operation of
various industries, so when GAI generates network information architecture optimization proposals, the
relevant personnel should take the lead in the network to avoid excessive technological dependence, and
at the same time, the monitoring of network security should be strengthened to implement the manual
synchronous monitoring.
51
www.leadleo.com
400-072-5588
Key findings
Generative AI assists the information and communications industry in breaking through the complexity of
traditional manual operation and maintenance management difficulties, driving the industry to intelligent network
mode transformation, enabling the multi-dimensional resources for joint optimization scheduling. In addition to
promoting the information and communications industry more accurately fit other industry sectors, expanding and
updating of the business model.
Source: Huawei, Sullivan.
Sullivan Market Research
2.8.1 Challenges and Developments in The ICT Industry
(1/2)
Industry's challenges and The New Form of “GenAI+ICT"
More broadly
More deepen
Newer
Data Type
Scope of
Application
ICT Product
Service
Evolution
AI
Digital
Robots
Digital
Synthesize
Digital
Source
Data
Corporations Industry chain Ecosphere All fields
ICT industry
Depth of
empowermen
t
ICT industry
Breadth of
empowermen
t
Digitize
Inspect
Operation
Visualization
Business Self-
Service
User
portray
Robots
Complementary
Diagnosis
Compliancy
Supervisory
Supply Chains
Synergistic
Internet
Platform
Credible Data
Space
Production
Cell
Simulation
Digitize
Supply Chains
Intelligent
Production of
Content Precision
Dynamic
Operations
Machine
Learning
Unmanned
Intelligent
Inspection
Genetic
Computationa
l Analysis
With the development of
generative AI-related technologies,
the ICT industry is more deeply
integrated with other traditional
industry businesses
Extension and expansion of
data and network, ICT industry
outward convergence ecology,
empowering the digital
transformation of thousands of
industries
The generation of new
technologies and new content
will further give rise to new
business forms and models in
various industries
❑Information and communications as a foundation to support the digital development of various industries, generative AI
technology, network operation mode to intelligent network conversion, driving the information and communications
industry more accurately fit other industry segments and business scenarios, and promoting the industry ecology's new
definition.
•Traditional industry challenges: With the network architecture and business ecosystem becoming increasingly
complex environments, the information and communications industry is faced with increased difficulty in manual
management and operation and maintenance, new upgrades to network facilities, service capacity transformation and
other challenges.
•New form of "Generative AI + information and communication": With the convergence and development of
generative AI technology, on the one hand, traditional information and communication is gradually transformed from
the original operation mode relying on experts' experience to the intelligent network mode, and the optimization
process is adjusted from the original post-trigger optimization process to the ex-ante prediction and active
optimization of performance to achieve timely adjustment and optimal allocation of network resources and
performance.
Chapter II: Compendium of Applied Practices
52
www.leadleo.com
400-072-5588
Key findings
Generative AI is mainly applied to the operation and security management in the communications
industry at this stage, which enhances the automation of network operation by transforming big data
resources into efficient network planning and fault diagnosis capabilities through intelligent algorithms.
Source: AI in Wireless Communication Systems, Sullivan
Generative AI Opportunity Points Graph
Sullivan Market Research
2.8.1 Challenges and Developments in The ICT Industry
(2/2)
Network Operation
Automatic Document
Generation
Automatic System Optimization Customer Service and Support Internet of Things
Applications
Promoting ecological convergence and innovation
Automatic Generation of
Security Solutions
Code Completion and
Optimization
Predicting Data Breaches
and Protection
Bug Detection and Fixing
Intrusion Detection and
Response
Automated Code Detection
Function
Intelligent Complaint
Handling
Collaborative Office System
User Behavior Analysis and
Optimization
Network Configuration
Automation
Operations
Network Security and
Protection
Improved network management efficiency and reduced
operating costs
Ensuring the security and stability of information and
communication systems
Note: Some applications involve multiple technical functions, and the corresponding colors represent the main functional technical modules.c
Web Optimization Proposal
Generation
Network Planning and
Deployment
Intelligent Base Station
Energy Saving
Channel State Information
Prediction
Base Station Load Balancing
Office Optimization
Intelligent Customer Service
Personalized Service
Recommendations
Real-time Analysis of Traffic
Data
Public Safety Monitoring
and Early Warning
Intelligent Real-name
Services
Intelligent Distribution and
Conservation of Energy
Smart City Management
Security
Management
Efficient scheduling and utilization of resources
❑As a vast connectivity network, the introduction of generative AI provides intelligent decision-making and optimization
capabilities, especially in the communications operation and security management scenarios, significantly improving the
industry's resource scheduling efficiency and service stability.
1Communications Operations: GAI can be used to generate planning and deployment of network operations and optimization recommendations to help
enterprises improve the efficiency of communications system construction and operation.
2Security Management: GAI monitors network operation status, identifies external anomalies and attacks, and encrypts data to ensure the security of
communication networks.
3Customer Service: GAI optimizes the quality of service by analyzing users' communication habits and using reinforcement learning algorithms to optimize the
allocation of resources and users.
4IoT application: Promote the comprehensive intelligence of communication networks in public utilities through real-time analysis and monitoring of IoT data
information.
User Resource Allocation
Optimization
Client Service IoT
Chapter II: Compendium of Applied Practices
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Information
collection
and
organization
Value
Created
Scenarios
Opportunity points
for generative AI
applications across
scenarios in the
ICT industry
53
www.leadleo.com
400-072-5588
Key findings
The deployment and optimization of communication networks involves the ecological operation of various
industries, so when GAI generates network information architecture optimization proposals, the relevant
personnel should take the lead in the network to avoid excessive technological dependence, and at the same
time, the monitoring of network security should be strengthened to implement the manual synchronous
monitoring.
Source: Ministry of Industry and Information Technology, Sullivan
Potential Application Risks
Technology Dependence and
Supply Chain Risk
Functional Areas
Risk of Technology Dependence: Big language modeling may output
factually incorrect information, rendering network troubleshooting
ineffective, and complete dependence on technology may bring risks to the
operation and security of the entire communication network.
Supply Chain Risk: The generative AI platform system may have some
cybersecurity risks, such as algorithmic backdoor embedding and code
security vulnerabilities.
Liability Attribution Risk
Network Dominance: Given the importance of timely resolution of network problems,
the relevant technical staff should take ownership of dealing with network problems,
supplementing the generative AI's recommendations with autonomous judgment.
Privacy and Data Security
Risks
Technology Risk
Attribution of Responsibility: The main parties responsible for cybersecurity attacks
and major risks such as operational failures due to generative AI miscalculations still
need to be further defined.
Risk of user Privacy Leakage: The communications industry designs a large
amount of user and sensitive data, which, in the event of a cyber-attack or
internal leakage, will result in the data being maliciously accessed or
transferred to an insecure environment.
Property Ownership: Innovative design of communication infrastructure, such as chips,
using generative AI requires consideration of how to obtain copyrights or patents and
protect the intellectual property rights of products already in production.
Risk of Data Misuse: Communication networks gather critical data from all
parties, and if generative AI lacks strict permission settings in data collection
and use, there may be data use beyond the scope of authorization and
leakage of relevant confidential information.
Technology Misuse: Generative AI may be used for phishing, large-scale
cyber fraud, and other illegal activities, making it more difficult to manage
security in the communications industry.
Cyberattacks: Attackers may reverse the use of generative AI to generate
attack code or improve the efficiency of cyberattacks, posing cybersecurity
threat.
2.8.2 Potential Application Risks in The ICT Industry
Network
Planning and
Deployments
Error Detection
and Fixes
Code
Completion and
Optimizing
Security
Solutions
Code
Completion and
Optimizing
Security
Solutions
Intrusion
Detection
Real-time
Analysis of
Traffic Data
Real-name
Service
Intellectualize
Personalized
Service
Testimonials
User Behavior
Analysis
Code
Completion and
Make Superior
User Behavior
Analysis
Error Detection
and Fixes
Sullivan Market Research Chapter II: Compendium of Applied Practices
54
www.leadleo.com
400-072-5588
Source: Alibaba Cloud, Sullivan
Lenovo Xiaotian
Application scenarios: ICT industry + AI personal assistant + device housekeeper + document processing +
intelligent creation, etc.
Core Evaluation Keywords
Model Lightweighting
and Adaptability
Security
Compliance
Capacity Expansion
Flexibility
Multilingualism and
Internalization
•Demand Adaptation: Alibaba Cloud's Xiaotian, built for
Lenovo based on its practical needs, helps customers and
users handle their work and life needs more conveniently.
•Data supply exclusivity: The solution provides Lenovo with
industry-specific high-quality dataset training and fine-
tuning services such as dialog scenarios.
•Scenario function generalization: Under this solution,
Lenovo‘s PC, cell phone and Pad product ends can realize
mutual expansion, and Lenovo can rapidly invocate
new functions through personal intelligences
and supports multilingual task processing.
•Controlled quality of generated content: Improve the quality of
generated content by testing model suitability based on a review
set and fine-tuning the model and optimizing the RAG.
•Security Compliance: In addition to the underlying regulatory
legal requirements, Alibaba Cloud provides Lenovo with a content
security API and an intervention module built into the modeling
platform to assist users in filtering inappropriate content from
model inputs and outputs by configuring keywords.
•Reasoneng efficiency optimization: Alibaba Cloud, in
addition to the adaptation of terminals and hardware,
enhances the inference efficiency of the scheme
through continuous optimisation of the inference
framework and model pressure testing.
•Reasoning and Deployment Cost Optimization:
The solution meets Lenovo's demand for performance
and cost optimisation in different business scenarios by
providing models of different sizes, which is highly economical
and cost-effective. In addition, Alibaba Cloud provides hundreds
Refined APIs for Lenovo‘s capacity expansion. No capacity limit.
•Training and Support Services: At the solution service support
level, Alibaba Cloud arranges regular communication activities for
Lenovo, synchronizes model progress with the customer side, and
cooperates to optimize the internal model. In addition, it provides
SLA 7x24-hour work order support and nail group support with a
minute response speed for cloud resources and large models
respectively.
•Scenario Value Satisfaction: Alibaba Cloud and
Lenovo have carried out in-depth cooperation,
based on the practical needs of Lenovo.
In addition, the company has developed a personal assistant
service on the end-side of the company's business, and has
jointly developed cooperation on artificial intelligence
terminals and enterprise intelligence solutions.
•Experience and customization satisfaction: Alibaba Cloud for Lenovo
targeted to enhance the optimization of generated content, and the
solution to achieve different end-side devices to expand each other, so
that the Lenovo Group internal sharing without barriers.
2.8.3 Best Practices in The ICT Industry - Lenovo Xiaotian
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Sullivan Market Research Chapter II: Compendium of Applied Practices
Appraisal
Scope
55
www.leadleo.com
400-072-5588
Source: Alibaba Cloud, Sullivan
Product Structure and Core Advantages
Solution Effectiveness
•User Experience Optimization: improving end-side smart interaction experience: the AI interaction experience of end-side devices has been
significantly improved through model lightweighting technology. Users are now able to enjoy smooth and efficient intelligent body services
such as AI Personal Assistant and Device Manager on their personal devices.
•Terminal Device Application Achievements: The end-side solution built by AliCloud for customers has successfully authorized 1 million PC
terminals and 350,000 cell phone terminals, and achieved a daily call volume of 100,000 times.
•End-side Model Requirements: Under the limited arithmetic
conditions of end-side devices, relatively small-sized models
are required to run efficiently, and customers have high
optimization requirements for model execution efficiency.
•Intelligent Product Solution Requirements: to enhance the
user experience of terminal devices and optimize the
workflow of enterprises, professional manufacturers need to
cooperate to create intelligent terminal devices and
enterprise intelligent body solutions.
•Model Lightweighting: AI models are lightweighted through model
distillation technology, enabling them to run efficiently on end-side
devices.
•Terminal Adaptation: Lenovo has cooperated with Aliyun for
targeted hardware adaptation to improve the execution efficiency
of the model on different chip platforms.
Family of foundational large models (Tongyi-Qianwen, etc.)
Generalized Large Model Training Tool
Model Center Application Center Data Center
Model
Training
Tuning
Mould
Evaluation
Mould
Deploym
ents
Model
Output
Settings
API Plug-
in
Manage
ment
Workflows
Lay out
PE
an
Engineering
project
Appliance
Evaluation Digital
Managemen
t
Digital
Depose
Mould
Indexing
Safety Protection
Light and Dark
Watermark
traceability
Output Recognition
Dynamic Interception
User Input
Risk Identification
Basic Information
Security
Qwen
Basic Model
End-side
Deployment
Qwen-Max
API Services
AI Terminal
Pre-assembled end-side model
AI PC AI phone AI flat panel AIoT devices
Lenovo Xiaotian Personal Intelligent Body
API
Encapsulation
Traffic
Management
Model ,anageme
nt
Data Analysis
API Interface API Interface
AI Personal
Assistant
Equipment
Steward
File Processing Game Assistant Personalized
Learning
Intelligent
Creation (religion) ...
Scenario-based AI Applications
Effectiveness of Implementation
Client’s Demands Solutions
Sullivan Market Research Chapter II: Compendium of Applied Practices
Alibaba Cloud and Lenovo Group jointly created the Lenovo Xiaotian and demonstrated
considerable user activity in authorized terminal products, achieving product personalized
services and strategic alignment and meeting Lenovo's demand for a "small and fine" end-side
model.
Alibaba Cloud Lenovo
56
www.leadleo.com
400-072-5588
Source: Alibaba Cloud, Sullivan
Recruitment Efficiency Assistant
Application Scenario: Information and Communication Technology Industry + Job Matching + Resume Optimization +
Interview Assistant
Core evaluation keywords
Recruitment Process
Optimization Accurate Job Matching Vertical Knowledge System Interview Efficiency
Improvement
2.8.3 Best Practices in The ICT Industry
- Recruitment Efficiency Assistant
•Requirement Adaptability: "Recruitment Efficiency Assistant"
provides a variety of intelligent functions for the recruitment
process, including resume optimization, resume screening,
recruitment assistant, etc., covering the core requirements of the
job search and recruitment vertical.
•Scenario Functionality Generalization:The solution helps
enterprises to improve the efficiency of resume screening,
enhance the accuracy of job matching, and at the
same time, helps job seekers to enhance their
competitiveness through intelligent resume
optimization services and improving the
quality of recruitment and job search.
•Generate Content Accuracy:With the support of the ZhipuAI GLM
series models, digital product innovation has been carried out to
create a data labeling system with extremely small granularity,
coupled with machine algorithms, deep learning and other
technologies, which effectively improves the accuracy of person-
post matching.
•In the AI mock interview scenario, the large model is
required to have strong information retrieval and
information matching capabilities to understand job
information and vertical domain knowledge. The
recruitment efficiency assistant makes the
recruitment process moreefficient and accurate,
significantly improving the technical
advantages of the platform.
•Deployment and Support Services: ZhiPuai’s
large model provides strong basic model technical
support for the innovation of the Zhilian Hiring AI
application, and its powerful language ability, multimodal
knowledge comprehension ability, and fast response speed, etc.,
all enable the product to quickly and accurately understand the
user's information, so as to bring high-quality services. Through
continuous optimization and iteration, recruitment efficiency
assistant has gradually improved its functional stability and user
experience.
•Highly Rated and Feedback from
Customers:The large model of Zhipuai
significantly improves the efficiency of job
matching and the accuracy of information collection,
helping Zhilian Hiring to maintain its leading position in the fierce
market competition. Both enterprises and job seekers commented
positively on "Recruitment Efficiency Assistant", saying that the tool
effectively simplifies the recruitment process and improves the
recruitment efficiency and job search experience.
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Sullivan Market Research Chapter II: Compendium of Applied Practices
Appraisal
Scope
57
www.leadleo.com
400-072-5588
•Zhilian Hiring's digital tools increase the efficiency of matching people and jobs by 50%, and the efficiency of companies harvesting resumes within 15
minutes of posting a job is increased by 25%.
•The AI resume change product has already helped job seekers write resumes with 100% higher efficiency, saving platform job seekers more than 20,000 hours
of resume writing time every day.
•The AI processing delivery assistant, on the other hand, utilizes the large model to generate dialogue capabilities and analysis capabilities, and the AI assistant
communicates with the candidates quickly and generates scores, cracking the problems of ineffective communication and limitations of professional
knowledge of HR in dealing with job submissions, and increasing the efficiency of processing submissions by 300%.
Source: Zhipuai, Sullivan
Product Structure and Core Advantages
Application layer
(computing)
+
Mesosphere
Model Layer
Hardware Layer
Enterprise + Job Seekers
2B Solutions
Information and communications technology
Industry
Large Models
Tools & Platforms
LLM0ps
Model
Platform
Model plug-ins, API, Fine-tune interface
Hardware
Optimization
GLM Series CodeGeeX Text-Embedding
Universal Large
Model
Arithmetic
Infrastructure Compute and Storage (laaS cloud vendors)
2C
Applications
Efficiency
class
emotional
needs
AI Governance
Model Security
Model Data
GPUs
Model
deployment Model
Training
•Inefficient Information Screening: ZhilianHiring Platform has huge
amount of job information and resumes, but the efficiency of
enterprises in screening resumes is low, and the recruitment process is
lengthy, which affects the recruitment effect. Enterprises urgently
need more accurate job matching tools to improve screening efficiency.
•Insufficient Competitiveness of Job Seekers: Job seekers' resumes lack
competitiveness, making it difficult for them to stand out. Zhilian Hirng
needs to provide value-added services for job seekers, such as resume
optimization and career planning, so that resumes are more in line
with HR's viewing needs.
•Enterprise: The Recruiting Assistant understands the needs of recruiters
through dialog, screens resumes for recruiters and makes targeted
recommendations. It also helps recruiters to analyze candidates' abilities
during the interview and gives comprehensive evaluation after the interview.
•Candidates: Job seekers enter their personal strengths and other information,
and recruitment efficiency assistant automatically generate a complete
resume, which can be optimized according to the needs of job seekers
resume.
Solution Effectiveness
Effectiveness of Implementation
Client’s Demands Solutions
Sullivan Market Research Chapter II: Compendium of Applied Practices
Using its powerful GLM series, Zhilian Hiring has developed the “Recruitment Improvement
Assistant”in cooperation with Zhipuai, aiming to optimize the recruitment process, improve the
experience of job seekers, and continue to develop the application of new artificial intelligence
technology in human resources services, leading the industry into a new era of digital intelligence.
ZhipuAI Zhilian Hiring
58
www.leadleo.com
400-072-5588
Challenges and Developments in
The Public Service Industry and
Best Practices
•Under the background of social structure imbalance such as demographic structure and
resource distribution, the public service industry is facing challenges such as slow response to
public information delivery, uneven allocation of public resources, and lack of precision in
service provision, etc. The introduction of generative AI assists the industry in improving
governance effectiveness and service quality, balancing the scheduling and allocation of
public resources, and promoting the development of the smart city model.
•At present, urban planning and management for generative AI in the public service in the key
areas of layout, generative AI for the public channels to establish data and information
interaction platform, through the public service object of the suggestion of the information
feedback reverse to improve the quality of service, optimize the allocation of resources and
planning decision-making, and promote the process of the smart city in a comprehensive
manner.
•When generative AI is applied to public service industries, potential risks such as public data
security, public acceptance, and administrative ethics need to be carefully managed to ensure
that references to new technologies are consistent with social values and ethical standards.
59
www.leadleo.com
400-072-5588
Key findings
Under the background of social structure imbalance such as demographic structure and resource distribution, the
public service industry is facing challenges such as slow response to public information delivery, uneven allocation
of public resources, and lack of precision in service provision, etc. The introduction of generative AI assists the
industry in improving governance effectiveness and service quality, balancing the scheduling and allocation of
public resources, and promoting the development of the smart city model.
Industry's challenges and The New Form of “GenAI+Public Service"
Chapter II. Compendium of applied practices
Sullivan Market Research
2.9.1 Challenges and Developments in The Public Service
Industry (1/2)
Source: The Risk Challenges of Embedding Artificial Intelligence in Public Service Governance, Sullivan
Challenges and New Opportunities
Traditional challenges in the public service sector
supply side
Population ageing: increasing supply pressures, failure of
intergenerational sharing mechanisms, "ageing before getting rich"
dragging down economic growth.
Demographic imbalance
Economic downturn and increased spending go hand in hand: the
contradiction between fiscal revenues and expenditures is becoming
more prominent.
conflict between income and expenditure
A single main body of supply: government-oriented, "unappealing", not
fully mobilizing social forces to participate in the process.
A single main body of supply
demand side
Excess supply of public services coexists with structural shortages: the
public's demand for the quality of public services is "on the rise".
Uneven distribution of resources
The concentration of population in large urban agglomerations poses
challenges to the provision of basic public services in large urban
agglomerations, such as difficulties in urban planning.
Population mobility and allocation conflicts
The diversification of channels for the expression of livelihood
demands in the Internet era requires a shift from homogenization to a
more precise and personalized service model.
Diversification of public needs
New forms of "generative AI + public services" Generative AI
Decision-making
capacity
Management
efficiency
Communication
and interaction Organizational
oversight
QOS
Decision evaluation
optimization
Data monitoring
and analysis
Future Scenario
Simulation
Routine task
processing
Management
Decision Support
Human resources
emancipation
Streamlining the
approval process
Efficient
personalized service
Innovative service
applications
Expanding Communication
Channels
Enhancing
organizational trust
Strengthening democratic
transparency
Strengthening
internal controls
Jumping out of the
corruption trap
Enhanced external
participation
❑As the traditional public service industry faces challenges such as the changing social structure and the uneven distribution
of public resources it brings, the industry's strategic use of generative AI has made significant progress at the levels of
decision improvement, service delivery, and performance management.
•Traditional challenges of the industry: The public service industry faces major challenges in optimizing service supply
and resource allocation, and the industry needs to balance the various supply and resource allocations in order to
achieve fairness and efficiency in public services.
•New form of "Generative AI+Public Service": The promotion and application of Generative AI in the public service
industry has helped the industry transform into a "holistic government model" and "intelligent service model".
Governments at all levels apply GenAI technology to build public service application scenarios, collect and analyze
cross-sectoral data, break data silos, strengthen government data governance, realize more accurate decision-making
and management, and improve service quality and efficiency.
60
www.leadleo.com
400-072-5588
Key findings
At present, urban planning and management for generative AI in the public service in the key areas of layout,
generative AI for the public channels to establish data and information interaction platform, through the public
service object of the suggestion of the information feedback reverse to improve the quality of service, optimize the
allocation of resources and planning decision-making, and promote the process of the smart city in a comprehensive
manner.
Source: Sullivan
Generative AI Opportunity Points Graph
❑Generative AI is gradually integrated into various operational aspects of the public service industry, of which urban
planning and management is the focus of the current social situation.
1
Data management and analysis: generative AI can efficiently integrate and analyze data in the public service industry, helping the
government to more accurately understand social needs and problems, and formulate more scientific and effective policies based on data
analysis results.
2Decision support and optimization: Generative AI assists the government in monitoring potential risks in real time and providing solutions
through the aggregation of data from various links. In addition, generative AI can assist government staff with repetitive tasks, such as
information queries, to improve efficiency and reduce labor costs.
3Public service and governance: Generative AI can link citizen needs with government service information to quickly create personalized
public service content and enhance citizen satisfaction. For example, the government can provide more personalized medical and health
services, education services, etc. to meet the diversified needs of citizens.
formulation
Law and order
department
Note: Some applications involve multiple technical functions, and the corresponding colors represent the main functional technical modules.
Policy Development and
Evaluation:
Security and
Emergency Service Social Security&
Welfare Services
Town Planning
and Management Civic
Development
Law Making and Analysis:
Emergency Planning:
Emergency Management:
Benefits and Support:
Health Data Monitoring:
Public Health Promotion:
Smart City Design:
Urban Smart Management:
Urban Renewal Projects:
Education and Career
Development:
Community Governance
and Participation:
Welfare Management System
Development
Formulation
and Regulations
Optimization of the allocation of public resources
Agile decision making, information efficiency across channels
Reduction of significant labor costs
Improving governance and public service capacity more generally
Chapter II. Compendium of applied practices
Sullivan Market Research
2.9.1 Challenges and Developments in The Public Service
Industry (2/2)
Draft Policy Generation
Modeling Policy Effects
Policy Consultation
and Q&A
Assessment of the
Effectiveness of Policy
Projections of Trends in
Policy Implementation
Extraction of Key
Information
Decision Support for the
Legislative Process
Automatic Generation of
Emergency Plans
Simulated Emergency
Response
Recommendations for
Optimizing the Allocation of
Emergency Resources
Real-time Monitoring and
Resource Scheduling
Risk Warning
Early Warning System
Development Assistant
Recommendations for
Preventive Measures
Benefits Resource
Planning Analysis
Real-time System Monitoring
Feedback
Analysis of the Integration
of Resource Requirements
Analysis of Disease Trends
Health Management System
Development
Smart Health
Education
Building a Knowledge Base
for Urban Planning
Smart City Planning
Recommendations
Urban Building
Layout Design
3D Reconstruction of
Historic Buildings
Real-time Monitoring of
Urban Traffic
Public Interactive
Platforms
Environmental 1uality
Monitoring
Optimization Analysis of
Environmental Indicators
Individualized Education
Services
Career Path Analysis
Counseling
Intelligent Vocational
Training Platform
Intelligent Community
Assistant
Intelligent Government
Platform
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Information
collection
and
organization
Value
Created
Scenarios
Opportunity points
for generative AI
applications across
scenarios in the
Public Service
industry
61
www.leadleo.com
400-072-5588
Key findings
When generative AI is applied to public service industries, potential risks such as public data security,
public acceptance, and administrative ethics need to be carefully managed to ensure that references to
new technologies are consistent with social values and ethical standards.
Source: Sullivan
Potential Application Risks
Regulatory and Ethical Risks Functional Areas
covered
Data Security and Privacy
Protection Risks
Technology Misuse and
Decision Dependency Risks
Issues of Public Trust and
Technological Autonomy
2.9.2 Potential Application Risks in The Public Service
Industry
Legal
Advice
Policy
Development
Ethical
Review
Personal
Privacy
Protectio
n
Data
Analysis
Network
Security
Decision
Support
Systems
Public Opinion
Monitoring
Risk
Assessment
Policy Lag: Existing regulations are lagging behind the rapid development of
generative AI technologies, leading to regulatory gaps.
Ethical and Moral Challenges: There is a need to form and update ethical standards
to address ethical and moral judgment in decision-making by AIGC.
Attribution of Moral Responsibility: when decisions or behaviors of generative AI are
controversial, the attribution of moral responsibility may be unclear.
Regulatory Harmonization: Addressing cross-sectoral and international policy and
regulatory harmonization to ensure global compatibility of technologies and policies.
Public
Relations
Management
Disclosure
of
Information
Data Security Vulnerability: sensitive data processed by generative AI may become
a target of cyber-attacks, leading to data leakage and privacy invasion issues, and
the need to strengthen the protection of sensitive data.
Misuse of Technology: Monitor and prevent inappropriate use of AIGC technology,
such as disinformation dissemination or manipulation of public opinion.
Lack of Trust: The public may be skeptical of AIGC-generated content, fearing its
accuracy and fairness. The public may lack an understanding of AIGC technology,
leading to information asymmetry and misunderstanding.
Privacy Violation Issues: Individuals' privacy may be inadvertently violated when
analyzing and generating content.
Problems of Data Misuse: Unauthorized data collection and use may lead to a
decline in public trust in government and public services.
Decision-Making Dependency: Governments and public service organizations may
become overly reliant on AIGC's analysis and recommendations, reducing their own
decision-making capacity and innovation.
Difficulty in Technology Regulation: As AIGC technology evolves, regulators may
face challenges in technology regulation to effectively monitor and control
technology adoption.
Differences in Acceptance: Differences in the acceptance of AIGC technologies by
different cultural or social groups may lead to uneven application.
Risk of loss of Autonomy: Over-reliance on the AIGC may lead to a loss of autonomy in
decision-making by government and public service organizations in certain
circumstances.
Sullivan Market Research Chapter II: Compendium of Applied Practices
62
www.leadleo.com
400-072-5588
2.9.3 Best Practices in The Public Service Industry
- Conchbots, trading robots, photonic engines and
Jarvis Vision Large Model Platform
Source: Alibaba Cloud, Sullivan
Application Scenario: Public Service Industry + Local Travel and Life Service + Code Assistant
Core Evaluation Keywords
Innovative Solutions for
Smart Mobility
Optimization of The Allocation of
Human Resources Security Compliance Highly Customized Program
Services
•Demand Suitability: Hello-Inc realizes the overall AI layout of
smart mobility through GenAI solutions, including Conch Robot,
Transaction Robot and Jarvis Vision Large Model Platform, etc.,
which meets the internal demand of improving internal efficiency
and customer experience and is in line with Hello-Inc's long-term
strategic goal of building smart mobility.
•Data Proprietary: The solution provides internal and external
industry pendant data to Hello-Inc for training, and makes use of
new industry data through RAG and SFT fine-tuning processing.
•Scenario Functionality Generalisation: Customers can
integrate RAG, plug-ins, etc. to build new applications
to meet different business scenarios based on the
orchestration capabilities of AliCloud's model
platform.
•Generated Content Accuracy: Conchbot has demonstrated high
accuracy and user adoption in roles such as office assistant and
code assistant, significantly improving business efficiency. In
addition, Alibaba Cloud continuously improves the quality of
content generation through SFT fine-tuning and RAG knowledge
base.
•Reasoning Efficiency Optimization: AliCloud designs optimized
reasoning solutions for Hello, including but not limited to
reasoning framework performance optimization, API and terminal
network optimization, etc.
•Security Compliance: Enabling Minword Filtering with
AliCloud Content Security API to achieve legal
compliance filing.
•Deployment Cost and Time Efficiency: Based on the
Alibaba Cloud Bailian Refinement Platform, it builds a
full-link model training tool to empower the programme
to achieve cost-effective optimisation, in addition, the programme
assists enterprises to optimise resource allocation through load
balancing and elastic capacity expansion.
•Training and Support Services: In addition to basic document
support, Alibaba Cloud provides customers with comprehensive
training tutorials covering solution training, large model
technology training, and other comprehensive training tutorials,
and realizes minute-level response in technical support.
•Scenario value Satisfaction: : Based on the
in-depth discussion and communication with
Hellobike, Alibaba Cloud designed a solution that
fully meets the expectations and significantly improves the
operational efficiency and user experience of the
customer's enterprise.
•Experience and Customized Satisfaction: The solution is designed
to cover two rounds: quality assurance and ancillary functions such
as digital pilots according to customer-specific usage scenarios,
continuously optimizing reasoning efficiency and internal system
integration articulation, and providing customer companies with a
full range of training tutorials and round-the-clock and timely
technical service support.
•Performance and Innovation Satisfaction: this cooperation
program is an innovative practice in the field of smart mobility.
Conchrobot, Trading Robots, Photon Engine, Jarvis Vision Big Model Platform
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Sullivan Market Research Chapter II: Compendium of Applied Practices
Appraisal
Scope
63
www.leadleo.com
400-072-5588
Source: Alibaba Cloud, Sullivan
Product Structure and Core Advantages - Jarvis Vision Large Model Platform
Solution Effectiveness
Reticulation Stockpile Comprehensive
Database Surety
Tongyi-Qianwen
-Max Tongyi Thousand
Questions-Plus
Tongyi-QianwenVL Qwen2-72B-instruct
...
Model
Training Model
Reasoning EAS Service
Monitoring
High Performance
Computing High Performance
Storage High-performance
Networks
Alibaba
Cloud
Alitongi Large Model AliCloud Artificial Intelligence Platform PAI
AliCloud Intelligent Computing Service PAI Lingjun
Hello Jarvis - Phantasmagoria Large Model
Platforms
Model Review Model Deployment Model Service
Monitoring Application
Programming Application Service
Inference Acceleration Framework with Optimal Benchmarks
Primary and secondary links + check-alive mechanism
One-click intelligent multi-card fine-tuning mechanism for the whole link
Intelligent control mechanism for self-developed quotas
Hellobike Local Travel and Life Service Business
Bicycles Scooter Coattails Take a Taxi Car Rental Financial Street Cat
•Intelligent Travel Layout: Hello-Inc aims to integrate new
technologies into local travel and life services to create
innovative "intelligent travel", and internally, it needs to
build an overall layout of AI for intelligent travel to
empower the whole chain of office, programming and
operation with AI.
•Optimize User Experience: We need to respond faster to
market demand and iterate product functions to bring
users a new experience that is more intimate and smarter,
and continue to drive intelligent travel with technology.
•Conch Robot: With multiple devices, multiple entrances, and full
compatibility with all types of document formats, it improves
internal efficiency and empowers the operation of the whole scene.
•Transaction Robot: directly empowering the business with Agent,
providing customer-oriented services, enhancing GMV and realizing
full process coverage.
•Photon Engine: Optimize marketing with AIGC technology and
improve efficiency of advertisement operation.
•Jarvis Phantom Vision Large Model Platform: rapidly incubate large
model applications to serve all of Hello Group's local mobility
businesses and accelerate the innovation process of smart mobility.
•Internal Efficiency: On the internal development side, the code assistant assists enterprises in improving the efficiency of code research and development by
12%+, and the accuracy rate of code completions reaches 80%; on the operation side, the ability to exist operation and maintenance personnel and the
efficiency of newcomer training are improved effectively through the frequent push of the Copilot mode of the large model, and the core business is covered by
93.1%, and the accuracy rate reaches 88%. On the product design side, Photon Engine assisted the enterprise in shortening the cycle of marketing graphics and
short video design and creation, in which the video production efficiency was improved by 10 times.
•User Experience Optimization: The empowerment of customer service assistants and transaction robots has significantly improved the user experience, as
evidenced by a 4.10% drop in the rate of customer complaints from hitchhiking vehicle owners and a 4.39% drop in the rate of customer complaints from
passengers, and for the adjudication of responsibility, the accuracy of hitchhiking vehicle adjudication of responsibility is as high as 87%+.
Effectiveness of Implementation
Client’s Demands Solutions
Sullivan Market Research Chapter II: Compendium of Applied Practices
Alibaba Cloud has joined hands with Hello-Inc to build an innovative program for intelligent travel
based on the Jarvis Phantom large model platform, assisting Hello-Inc to internally build the AI
layout of each link, which has significant benefits in improving the efficiency of the business
process as well as the user experience level.
Alibaba Cloud Hello-Inc
64
www.leadleo.com
400-072-5588
2.9.3 Best Practices in The Public Service Industry
- High-Speed Industry Large Model
Source: Baidu Cloud, Sullivan
Application Scenario: Public Service Industry + High-speed Intelligent Operation and
Management
Core evaluation keywords
Scene Customization Operational Intelligence Efficient Synergy of Information
from Multiple Sources
Risk Prevention and
Control
High-Speed Industry Large Model
•Scenario Model Specially Designed for Highways: The AI model
adopted by Jingxiong Expressway is the first one customized for
highway operation and management in the industry, which not
only reflects the high adaptability of the technology, but also
highlights the leading position of Jingxiong Expressway in the
industry in terms of innovation.
•Multi-dimensional Data Specialization: The program collects data
related to the operation of the highway industry on the basis of
Baidu's general large model capabilities conduct training.
•Industry-leading Technological Breakthroughs: The Jingxiong
Expressway project realizes technological innovations in the industry
and provides a brand-new solution for highway management.
•Improved Business Processing Capability: After the project
implementation, Jingxiong Expressway has realized high efficiency
and accuracy in business processing, reflecting the strong
performance of the technology
•Cross-business Instructions Directly: The solution opens
up the whole process link in the system for enterprises,
realizing that business instructions are directly delivered
to each link.
•One-stop Intelligent Management Platform: The Jingxiong
Expressway project provides an all-round intelligent solution
covering all aspects from data mining to business process
management, reflecting the vision of Jingxiong Expressway in
pursuing comprehensive intelligent management.
•Continuous Support from Baidu Intelligent Cloud: Baidu
Intelligent Cloud provided continuous technical support and
optimization services for the Jingxiong Expressway, ensuring the
long-term stable operation and continuous progress of the project.
•Efficiency Improvements in the User Centre:
the programme coordinates
Help Jingxiong high-speed project through
intelligent technology show
It has improved operational efficiency and user experience.
•Customer Recognized Intelligent Achievements: After the project
implementation, Jingxiong Expressway received positive feedback
from customers, who were highly satisfied with the intelligent
transformation results, which improved the overall command and
dispatching efficiency at the high-speed management and
operation levels.
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Sullivan Market Research Chapter II: Compendium of Applied Practices
Appraisal
Scope
65
www.leadleo.com
400-072-5588
Source: Baidu Cloud, Sullivan
Product Structure and Core Advantages
End
Point
Acqui
sition
Construction Data
Satellite Remote
Sensing Monitoring
Management Data
Drone Monitoring
Conservation Data
Video Surveillance
Operational Data
Smart Terminal
External Data
Other Sensors
Edge
Cloud
Centrali
zed
Cloud
Edge Application
Layer
Edge PAAS Layer
Edge IAAS Layer
Billing/Auditing Service Area Emergency Relief Supported Decision
Making Traffic Monitoring Structural Monitoring
Edge Service Center Edge Data Center Edge AI Services Edge IoT Services Video Cloud
Count Stockpile Reticulation Edge Computing
Cloud Server GPU Cloud
Server
Object Storage
BOS
File Storage
CFS
Load Balancing
BLB VPC Intelligent
Collaboration
Marginal
Autonomy
Central Application
Layer
Center PAAS
Layer
Center IAAS
Layer
Count Stockpile Reticulation Edge Computing
Cloud Server Object Storage
BOS
File Storage
CFS
GPU Cloud
Server
Load Balancing
BLB VPC Intelligent
Collaboration
Marginal
Autonomy
Operations Center Data Center AI Services Internet of Things (IoT)
Services
LBS Service Video Cloud
Large Visualization Screen Portal applet/app Open Service Platform
Billing/Auditing Road Network
Monitoring
Structural
Monitoring Concomitant
Information Service One Network
Connection
Supported Decision
Making
•Business System is Difficult to Manage Flexibly: The
customer's current business system exists in the content and
form of a single, solidified processes, information
fragmentation and other aspects of the difficulties resulting in
the lack of precision and flexibility of the enterprise digital
management and service decision-making, departments and
related personnel can not efficiently use the data,
collaborative work.
•High-speed Industry Large Model Program: Based on the highway
operation data for pre-training, it creates a high-speed industry large
model, with functions of intelligent monitoring, decision-making and
production, etc. It assists the operation business of the Jingxiong
Expressway in realizing the transformation from reactive response to
proactive discovery of problems, and comprehensively improves the
operation monitoring and accuracy rate of the Beijing-Hsiungary
Expressway network and the efficiency of emergency disposal and
the use of the business system.
•Personalized Demand Understanding: all the way to multi-party autonomous collaboration to achieve multi-system intelligent linkage, to
meet the different needs of the multi-end, for multi-party collaboration to reduce costs and increase efficiency, for high-speed
management to bring a more mature and more intelligent cooperative solutions, enhance the command and scheduling benefits, enhance
user satisfaction with travel.
•Improvement of Operation and Management Energy Efficiency: With the realization of high-speed operation and management
intelligence and automation in the whole process, the efficiency of enterprise business system will be improved by more than 80%, the
accuracy of event detection will be improved to more than 95%, and the efficiency of emergency response and information dissemination
will be improved by more than 80% according to the preliminary estimation.
Effectiveness of Implementation
Client’s Demands Solutions
Solution Effectiveness
Sullivan Market Research Chapter II: Compendium of Applied Practices
The highway industry big model solution targets the main difficulty of managing industry business
systems. It integrates four exclusive models including event detection, emergency response,
information processing and business interaction, and seven innovative applications,
comprehensively improving the operational efficiency of the Beijing-Xiongan Expressway and
achieving efficient collaboration among multiple parties along the way.
Baidu Cloud Jingxiong Expressway
66
www.leadleo.com
400-072-5588
Challenges and Developments in
The Automotive Industry and
Best Practices
•The reference of generative AI redefines the form of human-computer interaction, driving the
transformation of the traditional command interaction to active service mode, and optimizing
the driving experience. In addition, the integration of new technologies opens up the data of
the complete automotive chain, assisting automotive enterprises in perfecting the business
ecology, and creating the integration of "end-to-end" sensing and decision-making.
•Generative AI is reshaping the innovation landscape of the automotive industry, bringing
unprecedented efficiency gains and user experience optimization to automakers through
intelligent design, precision marketing and personalized services. In this transformation,
applications such as smart cockpits, road condition detection and safety monitoring have
become the frontiers of AI technology application.
•While the application of generative AI in the automotive industry shows great potential, it is
also has potential risks that cannot be ignored. Although enterprises are generally capable of
coping with the strategic and technological foundations of generative AI, they are less
confident in technology ethics, data governance and risk issues, which have become
significant barriers to AI adoption.
67
www.leadleo.com
400-072-5588
Key findings
The reference of generative AI redefines the form of human-computer interaction, driving the
transformation of the traditional command interaction to active service mode, and optimizing the driving
experience. In addition, the integration of new technologies opens up the data of the complete
automotive chain, assisting automotive enterprises in perfecting the business ecology, and creating the
integration of "end-to-end" sensing and decision-making.
Industry's challenges and The New Form of “GenAI+Automotive"
2.10.1 Challenges and Developments in The Automotive
Industry (1/2)
Pre-sales Vehicle
Experience Use
Product
Development Aftermarket
1Cogniti
vely 2Product
Research 3Purchase
Decision 4Purchase
Delivery
5Remote
Manage
ment of
Vehicles
6Vehicle
Occupancy
7Charging
and
Pull up
8Maintenance
9Client
Service After-sales Rights
and Support
10
1Product
Design 2Product
Development 3Testing and
Verification
Intelligent
Driving
Marketing
Service
After-sales
Service
Realization of "software-defined
automobile", vehicle-circuit collaboration,
more efficient use of data to empower
automotive products, accelerating iteration
Product Design
and Development
Generative AI improves the interaction
experience between the car and the driver
through speech control, and enhances the
driver's experience through autonomous driving.
Optimize vehicle performance design based
on market data analysis and leverage new
technologies for automated vehicle
assembly
Generative AI will replace most traditional
light maintenance tasks by generating
relevant repair assistance video tutorials
based on the driver's problem needs
❑The traditional automotive industry is facing challenges such as technological upgrading and reforming the 4S
model. The references of generative AI synergize the data network of the automotive industry chain, reshape
the relationship between automobiles and their users and car companies, and dramatically improve the ability
to understand automotive interactions.
•Industry Traditional Challenge: In the process of automotive industry's gradual transformation to digitalization, how car
companies can guarantee the effectiveness and qualification of data mining as well as the upgrading of transformation
technology has become a major challenge.
•The new form of "Generative AI + Automotive Intelligence": The integration of generative AI opens up the data
information of the complete chain from pre-purchase to after-sales maintenance, and assists automobile enterprises in
perfecting their business ecology by mining effective information in the generative chain to form an intelligent experience
of full-scene interconnection. In addition, generative AI redefines the form of human-computer interaction, transforming
the traditional command-based interaction into an active service mode, providing a driving experience with "emotional
value" for users.
Source: Sullivan
Sullivan Market Research Chapter II: Compendium of Applied Practices
68
www.leadleo.com
400-072-5588
Key findings
Generative AI is reshaping the innovation landscape of the automotive industry, bringing unprecedented
efficiency gains and user experience optimization to automakers through intelligent design, precision
marketing and personalized services. In this transformation, applications such as smart cockpits, road
condition detection and safety monitoring have become the frontiers of AI technology application.
Generative AI Opportunity Points Graph
Source: Sullivan
❑Generative AI is gradually penetrating into the various operational aspects of the automotive industry, in
which intelligent driving and safety, R&D and design, as well as marketing and service have become the
key areas for enterprises to apply generative AI.
2R&D and Design: In the design and R&D phase of automobiles, generative AI is able to integrate market demands, technical
parameters and user preferences to rapidly generate innovative automotive design solutions. It accelerates the transition
from concept to prototype, optimizes the design process and shortens time-to-market through virtual simulation testing.
3Marketing & Service: In marketing and service, generative AI leverages deep consumer insights to create personalized
marketing strategies and customized customer service experiences.
R&D & Design Markets & Services
Smart Cockpit:
Virtual Design Simulation
Personalized Driving Settings
Road Condition Detection
Vehicle Safety Monitoring:
Collision Warning System
Automotive Innovation Design:
Supply Chain Management:
Demand Forecast
Risk Assessment
Marketing and Analytics:
Marketing Content Generation
After-sales
service:
Shortening the development cycle
Reduce manufacturing labor costs
Improves driving safety and reduces the risk of accidents
Personalize the customer experience
Note: Some applications involve multiple technical functions, and the corresponding colors represent the main functional technical modules.
Production Process Optimization:
Predictive Maintenance
Customized Production
Customer Service and Support:
Real-time Traffic Monitoring
Environmental Sensing Systems
Emotion Recognition Assistant
Rapid Prototyping Iteration
Material Selection Optimization
Consumer Behavior Analysis
Brand Loyalty Analysis
Market Trend Forecast
Virtual Assistant Interaction
Automatic Emergency Braking
Intelligent Fault Diagnosis
Remote Vehicle Monitoring
Supply Chain Optimization
Parts Inventory
Management
Service Appointment Optimization
Intelligent Driving and Safety: Generative AI provides highly personalized safety warnings and driver assistance functions for
cars by analyzing traffic data and vehicle status in real time, significantly improving driving safety and driving experience. It
can predict potential risks and automatically adjust the vehicle response to realize intelligent risk avoidance.
1
2.10.1 Challenges and Developments in The Automotive
Industry (2/2)
Smart Driving and Safety
Sullivan Market Research Chapter II: Compendium of Applied Practices
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Information
collection
and
organization
Value
Created
Scenarios
Opportunity points
for generative AI
applications across
scenarios in the
Automotive
industry
69
www.leadleo.com
400-072-5588
Key findings
While the application of generative AI in the automotive industry shows great potential, it is also has
potential risks that cannot be ignored. Although enterprises are generally capable of coping with the strategic
and technological foundations of generative AI, they are less confident in technology ethics, data governance
and risk issues, which have become significant barriers to AI adoption.
Potential Application Risks
Data Security and Regulation Functional Areas
covered
Technology Maturity
Social Ethics and User Acceptance
Attribution of Responsibility
2.10.2 Potential Application Risks in The Automotive
Industry
Emotion
Recognition
Assistant
Personalized
Driving
Consumer
Behavior
Analyze
Data “on board”Complexity: For large model data and algorithms, China has put forward
classification and grading regulatory requirements. AI large model ‘on the car’, brought the
application of innovation, along with the diversity of enterprise data collection and the
exponential growth in the number of data collection, making the collection of data and analysis
of the use of more complex.
Technical limitations: generative AI may have limitations in areas such as mathematical
processing, and requires continuous technical iteration and optimization to meet the high
demands of the automotive industry.
Over-reliance on Technology: Generative AI models cannot achieve a complete understanding
of real-world situations, and relying entirely on technology to realize the overl driving process
may generate more serious risks, such as traffic accidents.
Ethical Review of Science and Technology: There is a preliminary framework for ethical review
of science and technology, but the question of what standards of review should be followed
for the different has yet to be resolved and discussed. One of the central ethical questions in
the activity of developing algorithms for autonomous driving, for example, is whether it is
permissible to quantitatively compare the value of life. For example, whether the lives of
multiple people are more valuable than the life of a single person, or whether the value of life
can be measured in terms of factors such as age, gender, and appearance.
Liability Determination Standard: Generative AI has the possibility of misjudging the driving
environment, which in turn leads to traffic accidents. In this context, the criteria for
determining the responsibility of the car, the manufacturer, the driver and the insurance
agency, as well as the claims process, need to be further refined and clarified.
Road
Monitoring
and Response
Automotive
Innovation
Design
Driving Data Confidentiality: Intelligent cockpit, personalized marketing and other parts
involve a large number of users' driving data and personal information, generative AI should
encrypt the storage of this part of the data, to prevent the leakage of personal data.
User Experience: Consider user acceptance and experience when promoting applications, and
avoid premature introduction of immature technologies to the market.
Vehicle
Safety
Monitoring
Automotive
Innovation
Design
Intelligent
Cabin
(electronics)
Intelligent
Cabin
(electronics)
Road
Monitoring
and Response
Vehicle
Safety
Monitoring
Source: Sullivan
Sullivan Market Research Chapter II: Compendium of Applied Practices
70
www.leadleo.com
400-072-5588
2.10.3 Best Practices in The Automotive Industry
- GPT-BI, GPT-Sales Assistant, GPT-Code, etc.
Application Scenario: Automotive Intelligent Manufacturing Industry + Sales Assistant + Code
Platform, etc.
Core evaluation keywords
GAI Reinvents Business
Processes Business Digital Twin Engineering Full Link
Optimization Security Compliance
•Demand Adaptability: The solution provided by Alibaba Cloud
meets the core demand of FAW to improve efficiency and GAI to
reshape its architecture, and products such as GPT-BI and Sales
Assistant comprehensively realize efficiency improvement and
architecture optimization in enterprise data analysis, consumer
reception, code development and other aspects respectively.
•Scenario Function Generalization: The first "business
unit“management concept drives the reshaping
of the overall process architecture, making
FAW's internal highly synergistic business
unit content to realize business digitisation.
•High-accuracy Data Analysis: The "GPT-BI" model ensures the
reliability of data analysis results with its 92.5% accuracy rate,
providing enterprises with precise decision support. In addition,
Alibaba Cloud continuously optimizes the quality of generated
content by actively collecting customer feedback.
•Reasoning Speed and Performance Optimization: The
solution significantly reduces response latency by
designing models of varying Q&A complexity
for inference, as well as optimising engineering
links.
•Deployment Efficiency and Cost Control: The
solution has a clear cost structure and provides a
flexible and scalable technology platform for client
companies, guaranteeing rapid deployment and
efficient operation of applications.
•Training and Support Services: The technical support team
provides rapid response through 7x24-hour work orders and nail
groups, ensuring minute-to-minute service support. In addition,
the user documentation system has been completely revamped to
cover all functions and operating steps, providing users with
comprehensive learning and usage resources.
•Scenario Value Satisfaction: Alibaba Cloud
continuously iterates and optimizes product
content based on FAW's needs and feedback,
specifically meeting customer needs.
•Experience and Customization Satisfaction: The solution has a high
degree of internal integration within FAW and provides
comprehensive, efficient and timely 24/7 technical support and
regular training communication within the customer.
•Satisfaction with Performance and Innovation: The accuracy of
the model-generated content fully met the client's expectations,
and FAW was provided with relevant considerations in terms of
data security compliance and inference latency optimization.
GPT-BI, GPT-Sales Assistant, GPT-Code, etc.
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Source: Alibaba Cloud, Sullivan
Sullivan Market Research Chapter II: Compendium of Applied Practices
Appraisal
Scope
71
www.leadleo.com
400-072-5588
Source: Alibaba Cloud, Sullivan
Product Structure and Core Advantages - China FAW GPT-BI
Appliance
(Dingtalk) BI Data Post-Processing Development BI Chart Visualization Rendering BI Chart Results Front-end Presentation
Big Data
&
Model Platform
Vector
Database
Look up
Reinforce Uantification
of Knowledge
Vector
Similarity
AI
Agent
BI Rewrites
Scene Model
Miniatures
Retrieval
Enhancement
BI Recall NL2SQL
Data Generalizatio Rejection Model FAQ Model
RAG Search Enhancement Module
Metadata
Quality
Standards
Bloodline
Data Security
Data Identification
Data Classification
Privilege Control
Data Auditing
Modeling R&D
Coding R&D
Resource
Account Implement
Priority
business Isolation
Dispatch Monitoring
Application Data Layer
Real-time
Data Acquisition
Batch
Data Collection
Semi/unstructured
Data Collection
Cloud
Infrastructure
Amenities Big Data Storage
Computation Engine Interactive Query Engine Multimodal
Computing Engine
Storage Base
Calculation Base GPU Computing Resources (Public Cloud)
AI Model Competency Center
AI Model
Training
R&D
AI Model
Deployments
Digital
Intend Digital
Annotate
Reviews
Construct Reviews
managerial
Mould
dePloyments
Inference
Service
Appliance
mAnagerial Appliance
Control
Mould
Evaluate
Mould
Analysis
Concern
Collects Concern
Optimizing
Model SFT Tuning
NLP Miniatures
Developing from 0
Alibaba Cloud Tongyi-Qianwen 14B
FAW-Big Data Competence Center
SecurityGovernance R&D
Data Lake
Storeys
Data Mart
Aggregation
Logical Entity
Sticky Data
Data Storage GPT-BI
Dataset
Development
Solution Effectiveness
•Improve Decision-Making Efficiency: The client requires a system that
provides real-time data insights and rapid decision-making support, as
well as improved efficiency in all aspects of business processes.
•GAI Reconstruction of Business Processes: FAW has always insisted on
technological innovation, and the introduction of GAI has driven the
demand for restructuring and integration of new technologies into
various business processes within the enterprise.
•GPT-BI: Finding data results in numerous BI reports and data reports, searching by
charts, and improving the efficiency of data analysis and decision-making within the
enterprise.
•GPT-Sales Assistant: subvert the traditional consultant reception mode, optimize
the consumer communication experience through human-computer interaction.
•GPT-Code: Assist development technicians to improve work efficiency and reduce
related labor costs.
•Business Process Efficiency Improvement: The products effectively help client companies improve quality and efficiency in data query analysis, sales conversion,
code development, product design and legal compliance review. Among them, GBT-BI model accuracy rate reached 92.5%, and GPT-Code assisted FAW R&D to
improve efficiency by 15%, of which AI code generation accounted for 20%.
•User Experience Optimization: The solution realizes multi-party user experience optimization. Within the enterprise, management shifts from decision-making
to selection with the assistance of GPT-BI, reducing a large number of man-hours at the level of information analysis and insight; sales, designers and IT
developers reduce repetitive work and stimulate creative inspiration with the support of GAI-enabled solutions. Outside the enterprise, with the support of
GPT-Sales Assistant, consumer reception and communication are more timely and efficient, and the store experience is greatly improved.
Data
Integration
AI Model
Test and
Evaluate
Effectiveness of Implementation
Client’s Demands Solutions
Sullivan Market Research Chapter II: Compendium of Applied Practices
Alibaba cloud and FAW have cooperated to create a series of solutions such as GPT-BI and GPT-
Sales Assistant based on cloud-native architecture, assisting the customer companies to realize
the improvement of quality and efficiency of various internal business processes, as well as
architectural remodeling under the convergence of GAI.
Alibaba Cloud China FAW
72
www.leadleo.com
400-072-5588
2.10.3 Best Practices in The Automotive Industry
- "1+6+N" Geely Hybrid Cloud Platform
Source: Baidu Cloud, Sullivan
Application Scenario: Automotive Industry + Intelligent Cockpit + Data Security Protection
Core Evaluation Keywords
Factory Digital Brain Efficient Collaboration of
Global Personnel
High Utilization of Resources
in All Segments High Degree of
Centralization
•Customized Solutions: In response to Geely’s internal needs, Baidu AI
Cloud create the "1+6+N" Geely Hybrid Cloud Platform solution,
which effectively meets Geely's needs for business efficiency,
personnel collaboration, and optimization of user experience.
•Intelligent Transformation with Strategic Alignment: the program is
fully aligned with Geely‘s digital strategy, reflecting the company’s
development in intelligent strategic foresight and deep
understanding of industry trends.
•Program Innovation: the solution focuses on consolidating the
cloud infrastructure, then moving existing businesses to the cloud
and data into the lake, building an artificial intelligence platform,
and maximizing the use and role of data in business operations,
forming continuous innovation and decision-making capabilities,
supporting business innovation and change in various fields, and
constantly giving rise to various new "business species."
•Flexible Application Deployment: Baidu's public
cloud is based on Intel Xeon scalable processors &
Intel Athlon SSDs, which can give full play to the
management advantages of proprietary cloud platform in
terms of performance, scalability, and availability, enabling rapid
system uptime and deployment.
•Real-time Service Upgrades: Geely's proprietary cloud platform
uses Baidu's architecture model from the bottom up, so it can
follow the development of Baidu's public cloud for real-time
upgrades, greatly reducing the difficulty of upgrading services at a
later stage.
"1+6+N" Geely Hybrid Cloud Platform
•Willingness to Cooperate in the Long Term:
Geely is willing to
Continued strategic cooperation with Baidu AI Cloud to
jointly accelerate the forward-looking exploration of
modern intelligent manufacturing.
•Customer-Recognized Intelligent Achievements: After the
project was implemented, Geely's internal business processes
were fully optimized and upgraded, and the digitalization of
automobile manufacturing continued to be upgraded; externally,
the large model of the intelligent cockpit provided Geely users
with a more interactive and convenient driving and riding
experience.
Appraisal
Scope
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Sullivan Market Research Chapter II: Compendium of Applied Practices
73
www.leadleo.com
400-072-5588
Source: Baidu Cloud, Sullivan
Product Structure and Core Advantages
•Demand for Intelligent Cockpit : With the rapid development
of AI, consumers have higher expectations for interaction
experience of intelligent cockpit, seeking richer ways of
human-vehicle interaction.
•Automotive R&D Data Management and Analysis: In the
process of automotive R&D, Geely faces the need to collect,
process, annotate and analyze a large amount of data, which
needs to be efficiently managed and utilized in order to
accelerate the R&D process.
•Cockpit Large Model: Baidu AI Cloud provides Geely with large model
technical support, realizing the operation of various functions of the
cockpit through the accurate understanding of driver and passenger's
intentions.
•"1+6+N" Geely Hybrid Cloud Platform: Based on the hybrid cloud
platform, Geely has constructed cloud facilities, cloud architecture,
security operations, development, and other capabilities to satisfy
Geely's internal data security protection and external rapid response to
user needs.
•Business Process Cost Reduction and Efficiency Improvement: Based on the “1+6+N” Geely Hybrid Cloud Platform, ZGH has achieved
cost reduction and efficiency improvement through unified architecture, centralized control, and reduction of duplicated construction, of
which the management and operation and maintenance costs have been reduced by 30%, and the efficiency of resource utilization has
been increased by 20%.
•Efficient Collaboration of Global Staff: Utilizing the cloud infrastructure provided by Baidu AI Cloud, Geely Automobile has established a
global business collaboration platform, which effectively solves the problems of timeliness and consistency of cross-regional collaboration.
R&D and design staff and suppliers worldwide are able to share information and work together in real time.
Operations and
Maintenance
Operations
Console
Operation and Maintenance Platform
Product O&M
General
Market Liabilities
Managerial Stockpiles
Managerial
Inspects
Managerial Surety
Managerial ...
Public Services and Basic Products
LAM Cloud
Monitoring Cloud Audit
Operating Platform
Tenant
Management Whitelisting Official CMS Discount
Management ...
Billing
Count Stockpile Reticulation Surety
GPUs
Cloud Server
Container
Service
Cloud
Server
Elastic
Telescoping
Integrated Layer
Server (computer)
Resources
Organization
Cloud Disk
CDS
File Storage CFS
Object
Storage
BOS
Local Disk
Load
Equalization VPC Special
Line
Elastic IP DNS
DDoS
Protection
Appliance
Firewalls
Key
Manageme
nt
KMS
Safety Check
SDR
Host
Security
Flow Audit
IDS
Database
Audits
Situational
Awareness
IPv6
Gateway
Middleware comprehensive database Big Data
Message
Queue
MQ
CNAP for
Microservice
Applications
API
Gateway
Efficiency
Cloud
MySQL DTS Mongo
DB TSDB PostgreSQL
SQL
Server DRDS SCS Oracle Cockroach
DB
Logging
Service
BLS
Elastic
Search
Data
Factory
Pingo
Data
Warehouse
Palo
Messaging
Service
Kafka
BMR Machine
Learning
BML
Data
Visualization
Sugar
AI
Speech
Recognition
Face
Recognition
Image
Recognition
Text
Recognition
Video
Analysis
Intelligent
Customer
Service
IoT
Physical
Access Object
Analysis Object-
Visible Location
Service
Rules
Engine Timing
Observation
Official Product
Console
Solution Effectiveness
Effectiveness of Implementation
Client’s Demands Solutions
Sullivan Market Research Chapter II: Compendium of Applied Practices
Baidu AI Cloud and Geely jointly created the "1+6+N" Geely Hybrid Cloud Platform, which not
only helps Geely achieve internal business process cost reduction and efficiency and global staff
efficient collaboration, but also opens up a brand new path for the digitalization of automobile
enterprises, i.e., to promote the digital transformation of the core business scenarios based on
the information pedestal to achieve the digital and intelligent business model.
Baidu Cloud GEELY
74
www.leadleo.com
400-072-5588
Challenges and Developments in
The Consumer and Retail
Industry and Best Practices
•Generative AI breaks the traditional commodity-oriented value chain, and through the
intelligent management and analysis of market data, the industry chain feedback and a series
of decision-making support are more in line with the market demand, realizing the
reconstruction of the value chain centered on the "user", and assisting enterprises in
responding to the market changes in a more rapid and sensitive manner.
•Generative AI is changing and empowering all aspects of the consumer industry, bringing a
series of value benefits to the consumer industry, such as product innovation, efficiency
improvement, marketing customization, and optimization of the user's consumption
experience, among which marketing and services are the key areas of the current generative
AI layout.
•Generative AI creates a more diverse connection between consumer and retail industry and
individual consumers, but at the same time the complexity and variety of consumer
information and the security of proprietary data within the organization become major
potential risks to consider.
75
www.leadleo.com
400-072-5588
Key findings
Generative AI breaks the traditional commodity-oriented value chain, and through the intelligent
management and analysis of market data, the industry chain feedback and a series of decision-making
support are more in line with the market demand, realizing the reconstruction of the value chain
centered on the "user", and assisting enterprises in responding to the market changes in a more rapid
and sensitive manner.
Source: Sullivan
Linear value chains centered on "commodities"
Traditional industry value chain and scenario challenges
2.11.1 Challenges and Developments in The Consumer and
Retail Industry (1/2)
Limited Product
Innovation
•Content innovation
breakthroughs are
limited
•Conceptualization
and testing cycle
constraints
•New product launch
progress is limited
Enterprise Data
Complexity
•Complexity of customer
information
•Complex data for
decision-making
•Complexity of
commodity data
•Complexity of data
sharing
Lack of In-depth
Market Research
•Missing market data
•Lack of information on
regional cultural and
economic differences
•Lack of competitive analysis
•Potential brand perception
deficit
Brand Marketing
Changes
•Changes in marketing
targets
•Changes in marketing
•Changes in marketing
objectives
•Changes in the
allocation of
marketing resources
Commodity
R&D Manufacturing Supply
Management Sales &
Distribution After-sales
Service
Services Cost
Increase
•Growing Demand
for Specialized
Talent
•Growing customer
demand for high
quality services
•Growing demand
for technology and
equipment iteration
"Generative AI + Consumer and
Retail"
User-centered value chain
reconstruction
Omni-Channel Convergence and
Data Management
•Generative AI breaks down
traditional online channel barriers,
such as virtual trials, to realize a
non-differentiated experience
between online and offline
•Intelligent data to optimize
business management, provide
decision support, and innovate
marketing campaigns that are
more in line with consumer needs
Intelligent Industry Ecology
Construction
•Generative AI applied to
all parts of the chain
•Collaborate to upgrade
the overall efficiency and
service of the industry
chain
•Market information feeds
the flow of the industry
chain
❑With the gradual integration of generative AI into all aspects of the consumer and retail industry,
the traditional linear value chain is being reconfigured to a new user-centric form.
•Accurate mining of market changes: The traditional linear value chain is commodity-oriented, with pain points
such as limited product innovation and complex enterprise data management, and is unable to respond quickly to
market demands and changes. The in-depth application of generative AI helps consumer and retail enterprises
capture consumption and demand with unprecedented precision, and accurate market analysis and insights feed
back to support enterprises in producing and designing products based on market trends and immediate demand,
forming a user-centered value chain.
Industry's challenges and The New Form of “GenAI+Consumer&Retail"
Sullivan Market Research Chapter II: Compendium of Applied Practices
76
www.leadleo.com
400-072-5588
Key findings
Generative AI is changing and empowering all aspects of the consumer industry, bringing a series of
value benefits to the consumer industry, such as product innovation, efficiency improvement, marketing
customization, and optimization of the user's consumption experience, among which marketing and
services are the key areas of the current generative AI layout.
Generative AI Opportunity Points Graph
Source: NVIDIA, Sullivan
❑Generative AI is gradually integrated into the various operational aspects of the consumer industry, in
which marketing and consumer services are the key areas for enterprises to use generative AI layout, and
is also the area where generative AI is currently more widely used.
1
Research and Innovation: Generative AI can integrate market and consumer information more comprehensively, assisting enterprises to
rapidly generate new product design ideas and solutions based on consumer demand and new market trends, and accelerating the process
of product innovation and testing.
2Management and talent Empowerment: Generative AI can summarize data from various links, monitor and identify abnormal data to help
enterprises screen potential risks in real time; in addition, generative AI can also assist highly skilled personnel in code writing and maintenance,
automatic generation of repetitive tasks, and so on, to help enterprises reduce labor costs.
3Marketing and Services: Marketing and services are the key consumer areas to integrate with generative AI, which is based on the
combination of commodities and consumer information to quickly create "one thousand people, one thousand faces" customized marketing
content and a new consumer service experience.
Management and Talent
Empowerment
Research and Innovation
Market
Research:
Market Information Synthesis
New Market Exploration and Analysis
Consumer Profiling and Behavioral
Analysis
Market Information Collection
Product design
innovation:
Product Design Assistant
Analysis of New Market Trends
New product
testing and launch:
Virtual Proof of Concept
Product Launch Decision Aid
Content Management:
Sensitive Data Monitoring
Information Risk Screening
Enterprise Data
Management
Channel Data Integration
Customer Information
Management
Talent empowerment:
Developer Code Assistant
Vocational Training Design
Automatic Generation of
Repetitive Work Content
Brand Marketing:
Marketing Content Generation
Personalized Content Customization
Marketing Effectiveness Simulation
Marketing Content
Compliance Verification
Sales
services:
Customer Service Assistant
Virtual Shopping Assistant
Product Virtual Trial
Virtual Model Showcase
Diversified and innovative
products to meet market needs
Agile decision making with information
efficiency across channels
Reduction of significant labor costs
More accurate and personalized product services, consumer experience optimization
2.11.1 Challenges and Developments in The Consumer and
Retail Industry (2/2)
Marketing & Services
Note: Some applications involve multiple technical functions, the corresponding colors represent the main functional modules.
Sullivan Market Research Chapter II: Compendium of Applied Practices
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Information
collection
and
organization
Value
Created
Scenarios
Opportunity points
for generative AI
applications across
scenarios in the
Consumer and
Retail industry
77
www.leadleo.com
400-072-5588
Key finding
Generative AI creates a more diverse connection between consumer and retail industry and individual
consumers, but at the same time the complexity and variety of consumer information and the security
of proprietary data within the organization become major potential risks to consider.
Source: Sullivan
Potential Application Risks
Reliability of Generated Content Functional Areas
covered
Mismatch between Virtual Space and Reality: Generative AI's simulation design based
on virtual space has the potential to be inconsistent with the real-world environment,
resulting in generated content that cannot be used in the real marketplace
Comprehension Bias: Generative AI has the potential for comprehension bias, so some
manual validation is required in, for example, program development and coding tasks
Liability Attribution Risk
Unclear Attribution of Responsibility: Responsible parties may include AI developers,
content providers, platform operators, etc. When a risk event occurs, unclear
attribution of responsibility may lead to disputes and losses
Privacy and Data Security Risks
Consumer Trust and Ethical Risks
Lack of Technical Standards: Due to the lack of harmonized technical standards, the
quality and performance of generative AI is difficult to assess, which may lead to risky
events
Product
Design
Risk of User Privacy Leakage: Generative AI requires a large amount of user data for
training and optimization, including sensitive information such as personal identity
information, shopping habits, interests and hobbies. Therefore, enterprises need to
consider the storage, transmission and isolation of such sensitive information.
Data
Management
Virtual
Product
Product Trial
Customer
Service
Assistant
Market
Research
Virtual
Shopping
AI Model
Showcase
Incomplete Laws and Regulations: The laws and regulations on generative AI are not
yet perfect, lacking a clear definition of responsibility and a mechanism for pursuing
responsibility
Code
Development
Risk of Proprietary Data Leakage: Generative AI models can involve proprietary
data and sensitive information such as product prices, profits, consumer information,
and negotiation strategies
Algorithmic Bias: Due to the limitations of data training and the inherent bias of
algorithms, generative AI may produce unfair or discriminatory recommendations,
and users need to understand the research and analytical methods behind the
generated content.
Misleading Information: Generative AI may produce misleading information during
the content generation process, misleading consumers into making wrong decisions
Code
Development
Decision
Support
2.11.2 Potential Application Risks in The Consumer and
Retail Industry
Marketing
Generation
Product
Design
Data
Management
Code
Development
Customer Service
Assistant
Market
Research
Data Information Delay: The output of content generation and design requires input
information and data with real-time characteristics, the time lag of input data exists in
the output results of low quality and inefficiency.
Sullivan Market Research Chapter II: Compendium of Applied Practices
78
www.leadleo.com
400-072-5588
Source: ZhipuAI, Sullivan
Sullivan Market Research
Health Guard Nutritionist "AI Mengmeng"
Application Scenarios: Consumer and Retail Industry + Consumer Health Assessment +
Nutrition and Health Content Generation and Q&A, etc.
Core Evaluation Keywords
Precipitation of
Specialised Data
Personalized
Nutritional Guidance Interactive Health
Advisor
Multimodal Human-
Computer Interaction
Appraisal
Scope
2.11.3 Best Practices in The Consumer and Retail Industry
- Health Guard Nutritionist "AI Mengmeng"
•Forward-looking Program: The "AI Mengmeng" built by Zhipu AI
for Mengniu is in line with Mengniu's long-term development
strategy from the perspective of better serving consumers.
•Data Exclusivity: "AI Mengmeng" uses Mengniu's private domain
knowledge related to nutrition and health accumulated for more
than 20 years, and combines the knowledge data from
cooperating nutrition and health authoritative organizations, and
has passed more than 10 nutrition and health certification exams
at home and abroad.
•High-performance Algorithm Optimization: "AI Mengmeng"
conducts in-depth analysis of user data to ensure its efficiency
and responsiveness in nutrition and health consultation.
•Innovative Multimodal Interaction Design: "AI Mengmeng"
provides an innovative multimodal interaction experience by
combining 3D visualization and natural language processing
technology.
•Customized Service: "AI Mengmeng"
•enables highly specialised and targeted services
through close integration with Mengniu's private domain
data and the knowledge of nutritional health authorities.
•Technology Integration and Compatibility: "AI Mengmeng" can
seamlessly interface with Mengniu's existing IT infrastructure and
data platforms, reducing technical friction and improving
implementation efficiency.
•Collection and Application of Satisfaction
Feedback: Mengniu gathers user experience
through multiple channels and suggestions, which in turn
optimize the service, creating a cycle of user-centered
continuous improvement.
•Interactivity and Educational Value: The 3D image and easy-to-
use interactive design of "AI Mengmeng" not only increase the
attractiveness of the service, but also provide an educational
experience.
Chapter II. Compilation of applied practices
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
79
www.leadleo.com
400-072-5588
Source: ZhipuAI, Sullivan
•Positive Feedback from Customers: Consumers enjoy more personalized and companionable professional nutritional health services, and
users' willingness to consume in both private and public domains has increased by more than 10%.
•Provision of Personalized Services: "AI Mengmeng" can provide hundreds of millions of consumers with personalized nutritional health
services, which include professional knowledge answers and personalized nutritional advice.
•Solve the problem of traditional operation mode: China's
professional nutrition and health service supply is seriously
insufficient (China has an average of only 0.3 dietitians per
100,000 people, far lower than the global level of 27 dietitians,
and the price is expensive). Demand for relevant professional
knowledge is growing, and more and more consumers and
families want to have a dietitian assistant who can provide
professional answers anytime, anywhere!
•AI Nutritionist Meng Meng: Smart Spectrum AI assisted Mengniu in
building an AI nutritionist Meng Meng based on MENGNIU.GPT, and
consumers can communicate with AI nutritionist Meng Meng
through natural language at any time of the day or night in 7*24H
to get expert-level personalized nutrition and health services.
"Core Advantages of AI Mengmeng
Specialized
Knowledge
Accumulation
Domestic Nutritional Health
Certification Examination
Registered
Dietitian Public
Dietitian
Secondary Health
Administrator
Chinese Medicine
Practitioner
Examination
China Internal
Medicine Physician
Exam
CAS
Counselor
Nutritional Dietitian
Theory Exam
Baby-sitter National
Rehabilitation
Physiotherapists
Postnatal
Rehabilitator Elderly
Caregiver
Tertiary Health
Administrator
International Nutritional Health
Certification Examination
Personal Fitness
Coach Certification
Sports
Dietitian
Certificate Physical
Training Specialist
Certification
American
Counselor Exam Korea National
Dietitians
Korea National
Nurse Practitioner
Japanese
Caregiver
U.S. CDM
Certified Meal
Manager
Japanese Mental
Health Social
Workers
Convenient
Interactive
Interface
Consumers scan
the QR code on
the bottle
Customized Onboarding
AI Planner
Target Setting
Simple and
Beautiful User
Interface
Product Structure and Core Advantages
Solution Effectiveness
Implementation
Plan
Evaluating Results
Developing a Plan
Data Analysis
Effectiveness of Implementation
Client’s DemandsSolutions
Sullivan Market Research Chapter II: Compendium of Applied Practices
By integrating Mengniu's profound nutritional and health data with cutting-edge AI technology,
"AI Mengmeng" provides users with personalized health advice and plans, and meanwhile
realizes the continuous iteration and optimization of services with the help of user feedback,
which promotes Mengniu's innovation in the field of digital health services.
ZhipuAI Mengniu
80
www.leadleo.com
400-072-5588
Source: iFLYTEK, Sullivan
, Sullivan
Vipshop's "12.8 Sale" Marketing Case
Application Scenario: Consumer & Retail Industry + Creative Marketing
Core evaluation keywords
Voice Interactive
Marketing Social Conversation to
Build Momentum
Full-scene Traffic
Guidance In-depth Implantation of
Brand
Appraisal
Scope
2.11.3 Best Practices in The Consumer and Retail Industry
- Vipshop's "12.8 Sale" Marketing Case
•Personalized Shopping Experience: The case enhances the
interaction with consumers through AI technology and provides a
personalized shopping experience, which is in line with Vipshop's
long-term development strategy.
•Market Adaptability: The marketing strategy successfully
attracted the target consumers,
•showing its high applicability in the
•e-commerce market and helping the brand
•to gain a competitive advantage.
•Technology Performance: The use of AI technology in marketing
campaigns, such as speech recognition and speech rate challenges,
demonstrates high-performance technology performance.
•Innovative Marketing Model: Innovative marketing
plays, such as voice input activation games and
smart cards, reflect Vipshop’s innovative
power in the application of technology.
•The Combination of Technology and Marketing:
The cooperation between Vipshop and Xunfei AI
Marketing Cloud ensured the smooth implementation of
the marketing campaign and demonstrated the close integration
of AI technology and marketing strategy.
•Popularity Guidance Strategy: The all-round popularity guidance
and cross-scene infusion strategy, supported by Xunfei AI
technology, maximizes the marketing effect.
•Immersive Shopping Experience: AI-
enabled marketing gameplay enhances the
customer experience user stickiness and
brand loyalty.
•Market Feedback Analysis: High arrival and replay rates, as well
as an increase in brand voice, reflected positive consumer
feedback on the campaign and provided the brand with market
insights.
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Sullivan Market Research Chapter II: Compendium of Applied Practices
81
www.leadleo.com
400-072-5588
Source:iFLYTEK, Sullivan
•Customized service: Customers seek a service experience that
can be tailored to their individual preferences and needs. They
expect companies to provide personalized product
recommendations, service options, and solutions.
•Technology Integration: Customers need technology solutions
that work seamlessly with existing life and work processes and
provide a one-stop shop, such as managing multiple business
processes through an integrated platform or automating the
home with smart home devices.
•AI-driven interactive activity: interactive game based on speech
recognition technology increased user stickiness through gamified
challenges, effectively enhancing the interactivity and fun of the brand
campaign.
•Cross-platform popularity guidance: Users can participate in the activity
by uttering a specific password in Xunfei Input Method. This solution not
only improves the visibility of the activity, but also shortens the user's
participation path and increases the conversion rate.
•High engagement and user stickiness: Vipshop successfully attracted a large number of users to participate through AI-driven interactive
activities. This innovative participation not only increased the attractiveness of the campaign, but also enhanced user stickiness by providing
instant feedback and rewards. During the campaign period, the H5 arrival rate reached 77.8% and the punch-out rate reached 51.9%.
•Brand Exposure and Volume Increase: Vipshop's "12.8 Sale" achieved a significant increase in brand exposure during the event, which not only
increased the online visibility of the brand, but also further expanded the brand's reputation through the social fission effect.
Product Structure and Core Advantages
Solution Effectiveness
Event Showcase Highlights of the program of activities
Game Trigger, Creative
Marketing
•Vipshop joined hands with iFLYTEK to create a speed of
speech challenge, using social and entertainment as an
entry point to spread the brand campaign.
Popularity
Aggregation,
Increase
Conversions
•Xunfei AI Marketing Cloud and Xunfei Input Method have
created a creative marketing game, Voice Game, to
customize exclusive communication ideas for the brand
and implant in-depth brand information, including
customized skins for the voice input interface, brand
activities, etc., to create an immersive experience for the
consumers through game interactions and to deepen the
dissemination of the brand activities.
Product
Solution Core
Advantages
Perfect Product
Setup Digital Skill Originality Take
A product system
that meets the
multiple needs of
consumer insights,
brand building and
potential customer
mining
With a massive
data base, the
whole scenario is
fused and
integrated to guide
the optimization of
marketing
decisions.
Relying on AI
and big data
capabilities to
comprehensivel
y improve
marketing
effectiveness
Integrate high-
quality online
communication
resources to
reach target
users efficiently
in the whole
scene.
Applying new
technologies to
create
innovative new
forms of
advertising to
meet diversified
communication
needs
Effectiveness of Implementation
Client’s Demands Solutions
Sullivan Market Research Chapter II: Compendium of Applied Practices
Vipshop’s “12.8 Sale” marketing case realized efficient interaction and personalized experience
with consumers through the application of AI technology, and meanwhile, with the help of
innovative marketing means, such as voice input and social fission, it effectively enhanced brand
exposure and user participation, showing the practical value and innovative potential of AI in e-
commerce marketing.
iFLYTEK Vipshop
82
www.leadleo.com
400-072-5588
Challenges and Developments in
The Education Industry and Best
Practices
•The development of generative AI comprehensively empowers the optimization of the
"five factors" of the education industry, upgrading the quality of education and
optimizing the allocation of educational resources by innovative means, and boosting
the education industry to transform into three major directions: intelligent course
content, innovative teaching methods, and upgrading and optimization of the
educational environment.
•Generative AI is changing and empowering all aspects of the education industry,
bringing a series of value benefits to the education industry such as industrial
upgrading, teaching innovation, organizational improvement, etc.Thereof, teaching
content and organizational forms are the key areas of generative AI layout.
•While generative AI has spawned innovative models in the education industry, it has
also posed challenges. These include the quality of educational content, the potential
risks to the privacy and security of educational subjects, the recognition of educational
equity and responsibility, and the need to build public trust and address ethical risks.
83
www.leadleo.com
400-072-5588
The development of generative AI comprehensively empowers the optimization of the "five factors" of
the education industry, upgrading the quality of education and optimizing the allocation of educational
resources by innovative means, and boosting the education industry to transform into three major
directions: intelligent course content, innovative teaching methods, and upgrading and optimization of
the educational environment.
Source: SenseTime, Sullivan
Industry's challenges and The New Form of “GenAI+Education"
The "Five Factors" of Generative AI Enabling the Education Industry
Teaching Principal Part
AI assist teachers in reducing complex tasks
LLM
Assist teachers in automatically generating lesson
plans and accompanying exercises based on
subject and history content.
Prepare
Lessons
Pedagogical
Assessment
Assist teachers with lesson plan generation and
provide individualized instructional programs
Evaluate students’ uploaded assignments and
generate student learning analysis assessments
Deepen
Knowledge
Interdisciplinary
Teaching
Information and knowledge from a large language
modeling corpus assists teachers in deepening
their own learning
Assisting teachers in realizing the design
generation of interdisciplinary instructional
programs
Shift to individual
training
Focus on
Comprehensive
Assessment
AI assist students in optimizing learning experience
Interactivity Provide students with personalized analytical
guidance and heuristic interactions
Learning
Outcomes Automatic generation of learning plans and
learning scores, comprehensive control of
learning results
Shift Flexible
Interaction
Shifting
Convenient
Learning
Fully Mastered
Knowledge Points
Teaching Methods Educational Content
•Intelligent upgrading of the
traditional teaching environment to
enhance the learning and teaching
convenience of the teaching body
and to meet diversified teaching
needs.
•Extending and expanding the
functional boundaries of traditional
educational environments and
facilities
•Enhance the interactivity and fun
of the teaching process, and
increase students' participation
and initiative in the course.
•Innovative forms of teaching
organization and course structure
to promote practice and
innovation in interdisciplinary
projects
•Rapid production and updating
of course content to improve
teaching efficiency
•Personalize course content to
enhance learning autonomy
and flexibility, taking into
account the dynamics of the
social environment and the
situation of the students.
❑The education industry is facing a series of challenges such as uneven distribution of resources and old-
fashioned teaching system in its development, and the integration of generative AI empowers the education
industry to transform in three major directions: intelligent course content, innovative teaching methods and
upgraded educational environment.
•Traditional Industry Challenges: The education industry faces major challenges such as uneven distribution of resources,
solidification of the education system, and low teaching efficiency.
•"Generative AI+Education" New Format: Generative AI is comprehensively empowered by five factors in the education industry,
boosting the transformation of the education industry in the three major directions of intelligent course content, innovative
teaching methods and upgraded and optimized education environment.
2.12.1 Challenges and Developments in The Education
Industry (1/2)
Teaching
Goal
Key Findings
Educational Environment
Sullivan Market Research Chapter II: Compendium of Applied Practices
84
www.leadleo.com
400-072-5588
Key findings
Generative AI is changing and empowering all aspects of the education industry, bringing a series of
value benefits to the education industry such as industrial upgrading, teaching innovation, organizational
improvement, etc.Thereof, teaching content and organizational forms are the key areas of generative AI
layout.
Generative AI Opportunity Points Graph
1Teaching content and channels: Generative AI automates the process of learning material generation, teaching video creation, coursework
correction, and course progress control based on the needs of learning, so as to promote the growth of excellence.
2Teaching organization and methodology: Generative AI enables pedagogical interaction enhancement through interactive path simulation,
course material generation, and virtual teacher interaction; and innovation through cross-disciplinary knowledge integration, question
answering, and data analysis.
3Educational environment and scenarios: Generative AI rapidly produces an optimised solution for the educational environment, achieving
efficient use of existing assets, transforming hardware equipment, and improving the quality and efficiency of the teaching process; it is
capable of rapidly integrating high-quality resources, constructing innovative learning platforms, and optimising the experience.
Teaching Content and
Channels Educational Environments
and Scenarios
Supplemental Instruction:
Automatic Syllabus Generation
Enhanced Interaction:
Empowering the transformation
Educational aids to promote innovation in the form of teaching content
Enhance pedagogical interaction and creative integration of disciplines to improve quality and effectiveness
Transforming the educational environment, expanding the boundaries, and enhancing education experience
❑Generative AI is gradually integrated into the whole chain of education, providing technical support for
the integration of online /offline teaching and the strengthening software and hardware equipment.
Verificate and Correct Coursework
Assisted Retrieval Teaching Materials
Multimedia Courseware Outputs
Personalization:
Customized Learning Materials
Monitor and Adjust Courses
Digital Teacher:
Online Digital Teaching
Batch Video Content Creation
Virtual Teacher Interaction
Interactive Path Simulation
Discussion Material Generation
Creative Ideas Aid Sorting
Interdisciplinary Data Visualization
Disciplinary Innovation:
Interdisciplinary Knowledge Integration
Cross-disciplinary Questions Answering
Assist Innovative Projects
Facility Remodeling:
Hardware Equipment Transformation
Digital Asset Verification
Academic Decision Support
Innovation in Campus Environment
Boundary Extension:
Dissemination of Quality Resources
Lifelong Learning Platform
Anthropomorphic Counseling
Online Learning Community
2.12.1 Challenges and Developments in The Education
Industry (2/2)
Teaching Organization
and Methodology
Note: Some applications involve multiple technical functions, the corresponding colors represent the main functional modules.
Source: Tsinghua University, Sullivan
Sullivan Market Research Chapter II: Compendium of Applied Practices
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Information
collection
and
organization
Value
Created
Scenarios
Opportunity points
for generative AI
applications across
scenarios in the
Education
industry
85
www.leadleo.com
400-072-5588
Key Findings
While generative AI has spawned innovative models in the education industry, it has also posed
challenges. These include the quality of educational content, the potential risks to the privacy and
security of educational subjects, the recognition of educational equity and responsibility, and the need
to build public trust and address ethical risks.
The Reliability of Generated Content
Variable Quality of Educational Data: The unstructured nature of
educational data presents that the content generated by generative AI may
be incorrect and false, potentially leading to false information overload
.
Intelligent
Lessons
Technological Backdoor: The opaque nature of the logic algorithm raise
concerns. Additionally, there is a risk of generative AI becoming a tool for
different groups to compete for educational resources.
Delayed Data Information: Generative AI input data is delayed, resulting in
poor timeliness of output results and reduced educational value
Virtual
Teacher
Assisted
Marking
Privacy and Data Security
Privacy Infringement in Education: Generative AI creates a comprehensive
digital surveillance system that collects a large amount of personal data in
interactive education, which may lead to privacy breaches.
Fairness and Responsibility
Educating Trust and Ethics
Data Security Risks: As generative AI becomes fully integrated into the
education system, it may excessively collect and utilise user privacy data,
and illegally export privacy data during interactive learning, posing a data
security risks.
Regional Injustice: Generative AI is heavily reliant on the development of
science and technology, which may lead to an imbalance in the regional
distribution of educational resources.
Differentiation of Outcomes: The degree of mastering generative AI will
further differentiate the allocation of educational resources, leading to
differentiation in the long-term development of individuals.
Judgment of Responsibility: Generative AI may generate incorrect
educational materials, which could lead to serious educational accidents.
Ideological Hardening: Users may develop an illusion of intellectual
authority in the absence of critical thinking, leading to information cocoon.
Social Relationships: Generative AI weakens the teacher-student
relationship, leading to emotional indifference; while overemphasising the
human-machine relationship can lead to over-reliance and alienation.
Black Box Problem: Generative AI may misrepresent reality during its
application, exacerbating the uncertainty and trust crisis in educational
governance.
Educational
Materials
Virtual
Interaction
New
Hardware
Personate
Tutoring
Teaching
Assistant
Educational
Materials
Learning
Platform
Extended
Classroom
Educational
Materials
Answers
Residency
Personate
Tutoring
Personate
Tutoring
Potential Application Risks
2.12.2 Potential Application Risks in The Education
Industry
Source: Sullivan
Sullivan Market Research Chapter II: Compendium of Applied Practices
86
www.leadleo.com
400-072-5588
2.12.3 Best Practices in The Automotive Industry
- SJTU AI for Science Open Source Platform
Source: Baidu AI Cloud, Sullivan
Application Scenario: Education Industry + Research Platform
Core evaluation keywords
Research
utomation Arithmetic
Optimization
Data Security
Protection
Educational
Experience
Appraise
scope
SJTU AI for Science Open Source Platform
•Research Process Automation: The AI for Science platform greatly
improves research efficiency by automating tools and processes,
reducing researchers' time and energy consumption on data
retrieval and experimental validation.
•Research Data Management: The platform realizes integrated
management and development of research data for federal
modeling and federation of multidisciplinary data.
•AI Research Paradigm Innovation: The platform realize the
combination of generative AI and scientific research, which and
provides new perspectives and promotes the innovation of
scientific research methodology.
•High-performance Arithmetic Support: Baidu AI
Cloud provides SJTU researchers with a
strong arithmetic capacity to accelerate
model training and data processing.
•Rapid Platform Iteration: The AI for
Science platform is capable of rapid iteration
and upgrades to adapt to the changing needs of
research, ensuring the technology remains front
•All-round Technical Support: Baidu AI Cloud provides all-round
technical support, including data center and AI center, to ensure
the stable operation of the platform and the smooth going of
scientific research.
•Remarkable scientific research results:
The application of the AI for Science platform
has brought a series of pioneering scientific
research results to Shanghai Jiao Tong University,
enhancing the school's academic status and influence.
•Positive user feedback: The platform’s use experience has
received positive feedback from scientific researchers, which has
improved the satisfaction and efficiency of scientific research
work and promoted the transformation of scientific research
results.
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Sullivan Market Research Chapter II: Compendium of Applied Practices
87
www.leadleo.com
400-072-5588
Product Structure and Core Advantages
Research Ecology
UNPOS
Intermodal Events
Fund Support
Joint Research
Resource
Integration
Curriculum, Research,
Training
Technology Salon
Contributor Community
Scientific Data Center
Catalogue and Parse
Scientific data maps and
Data quality control
Multimodal Data Lake
PB-level multimodal
data storage
Heterogeneous
Computing Resource
Deep learning of models
Data Multi-Cloud
Collaboration
Compliance Audit
Artificial
Intelligence
Platform
Low-code AI
Data Analyze of Scientific
Self-assembly
AI Solver
Strategic Decision of
Scientific Experiments
Generative AI
Scientific data generation
Meta-universe Simulator
Intuitive Deduction of
Scientific Phenomenon
AI + Material Science
Experiment Design and
Regular Extraction
AI + Molecular Science
Chemical Reaction and
Precision Synthesis
AI+Fluid Simulation
Simulation and Inversion
of Fluid Phenomena
AI + Urban Science
Large-scale systems
Behavioral simulation
Artificial Intelligence Platform
Scientific Collaborative Platforms
Source: Baidu Cloud, Sullivan
•Baidu's self-developed ERNIE Bot serves as the foundation for the
SJTU AI for Scientific Data open-source platform. It incorporates a
comprehensive range of Baidu Intelligent Cloud capabilities,
including a big model, AI center, data center, and other core
competencies, to facilitate the integration of generative AI into
scientific research scenarios.
•Built the first AI for Science research platform in domestic university, creating an innovative paradigm that combines generative AI
technology with research scenarios.
•Improve research efficiency, liberate productivity, and release a series of first-of-its-kind AI4S research results.
•Development of the Baiyulan Science Large Model - Chemical Synthesis (BAI-Chem 2.0) Big Model.
•Upgrade Big Model 2.0 and co-research Law (BAI-Law-13B) Large Model.
•Published the main results of AI+ urban science on the cover of Nature Computational Science.
•In light of the significant advancements and maturity of
generative AI, SJTU is pursuing the establishment of an open
AI4S R&D platform. This aims to leverage the capabilities of AI
to enhance the efficiency and quality of scientific research,
while promoting the in-depth application.
Solution Effectiveness
Effectiveness of Implementation
Client’s Demands Solutions
Sullivan Market Research Chapter II: Compendium of Applied Practices
The AI4S scientific data open platform is first new paradigm application for reconstructing
scientific research based on generative AI Large Model in China.
Baidu Cloud SJTU
88
www.leadleo.com
400-072-5588
2.12.3 Best Practices in The Education Industry
- Student Development System, etc.
Source: iFLYTEK, Sullivan
Application Scenario: Education Industry + Intelligent Education Platform
Core evaluation keywords
Digital
Management Individualized Instruction Integration of Industry,
Academia and Research Smart Education Ecology
Appraise
scope
Student Development System, etc.
•Personalized Educational Support: The career guidance system
provides students with personalized learning paths and career
development advice, following the trend of personalized
education. It also adapts to the new college entrance examination
reform‘s demand for diversified choices for students.
•Strategic Development Consistency: The system's design concept
is dedicated to better serving students and promoting their
comprehensive development through advanced technological
means.
•Application of Advanced Technology: The system utilizes current
advanced big data and AI technology to conduct comprehensive
analysis of students' subject interests and potential, offering
scientific guidance on subject selection.
•Innovation in Education Model: Implementation of the career
guidance system has driven innovation in the education model,
providing students with a more flexible learning experience,
which
is challenging to achieve in the traditional education model.
•System Integration and Implementation:
The efficient collaboration during the
implementation of the system has ensured the smooth
deployment of system and integration with existing
educational processes
•Ongoing Technical and Service Support: Continuous technical
support and service updates provided by iFLYTEK guarantee that
the system can continue to meet the needs of school
development and adapt to the latest trends in education
technology.
•Intuitive Interface Design: The system is
designed to be easily used, lowering the
threshold for students and enabling them to make full use
of the various functions.
•Positive User Feedback: By collecting and analysing feedback
from students, parents and teachers, the system has achieved
positive results in improving user satisfaction, which also provides
valuable suggestions for continuous improvement.
Capacity Analysis
Functionality and Adaptability Performance and Innovation
Deployment and Support Experience and Satisfaction
Sullivan Market Research Chapter II: Compendium of Applied Practices
89
www.leadleo.com
400-072-5588
Source: iFLYTEK, Sullivan
The cooperation embodies an innovative practice of smart education, which promotes the
intelligent and precise delivery of educational services through the personalized career guidance
system. The system not only improves the efficiency of teaching management, but also optimizes
the learning experience of students.
•Personalized Learning Needs: Students expect to have
learning experiences that match their individual interests
and abilities.
•Enhance management efficiency: Educators seek more
efficient ways of managing to accommodate complex
teaching arrangements.
•Technical support and services: Schools need to ensure the
stable operation of the system while receiving timely
technical support.
•Personalized Learning Path Planning: The System customizes
learning plans for each student by analyzing their learning habits,
performance and interests.
•Integrated Management Platform:The system provides an
integrated management platform which includes class scheduling,
electronic class cards etc. to simplify management process and
enhance the efficiency. Ongoing technical service : iFLYTEK provides
system maintenance and rapid response services to ensure smooth
operation .
•Personalized Education Results: The school has successfully achieved personalized education, effectively promoting the development of
students' potential.
•Improved Management Efficiency: The system have greatly improved the efficiency of teaching management, reducing the burden on
teachers and making the allocation of teaching resources more efficient.
•Improved Learning Experience: The system's timely feedback and personalized guidance make students more proactive in their learning,
giving them a richer learning experience and improving user satisfaction.
Solution Effectiveness
Product Structure and Core Advantages
Based on Xunfei Intelligent Education
Serve Different Educational Roles
Educational administrators
•Scientific Management: Enhancing
management efficiency and promoting
regional education development
School Administrators
•Lifelong Learning: Enhancing the quality of
school teaching and learning, promoting
the All-round Development of Students
Principals
•Reducing workload : Reducing teachers'
repetitive work and improving teaching
quality
Head of a Household
•Home-school co-education: Solve the
problem of home tutoring, understand the
situation of children in school
Schoolchildren
•Self-directed Learning: Enhancing academic
standards and Promoting all-round quality.
Covering Whole Education Scene
Regional Education Governance
•Teaching individually and integrate planning
to promote the digital transformation of
education
Campus Position Building
•Reducing burden and increasing efficiency,
teaching according to students’abilities,
cultivating morality and nurturing people,
and promoting the all-round quality.
Independent Study
•Evaluate students' learning situations, boost
confidence through results, and cultivate the
ability to learn independently.
Smart Exam
•Artificial Intelligence helps make China’s
education exams Safe, Efficient, Fair and
Equitable.
Effectiveness of Implementation
Client’s Demands Solutions
Sullivan Market Research Chapter II: Compendium of Applied Practices
iFLYTEK High School Attached to ZJU
90
www.leadleo.com
400-072-5588
Challenges and Developments in
The Enterprise Application
Industry and Best Practices
•Since the uneven digitalization of enterprises, the enterprise application industry faces
challenges such as difficulty in synchronizing office system software and weak system
integration capabilities. The emergence of generative AI can help the industry to connect
internal enterprise information data based on advanced technologies such as knowledge
graphs and natural language processing, efficiently extract effective decision-making
information, and generate personalized office systems.
•Generative AI empowers enterprise users with its data-driven decision-making capabilities,
enabling office collaboration, and automation. It integrates the flow of internal information
and key decision-making information support, and upgrades and develops personalized
products based on the analysis of users' office behaviors, creating a new way of working.
•Generative AI share the risk of enterprise data and information leakage, as well as the key link
of the data lag and technical failures leading to system platforms throughout the chain of
information flow paralysis. Therefore, enterprise users and application developers need to add
supervision and the relevant level of development technology.
91
www.leadleo.com
400-072-5588
Industry's challenges and The New Form of “GenAI+Enterprise Application"
Source: Sullivan
Industry Traditional
Challenges
Generative AI
Empowerment
Low level of Digitization
01
The account between different
collaborative office software is not
synchronised, so that the business is not
interoperable, data sharing is difficult,
resulting in the problem of office silos.
Extracting Unstructured Data
02 Plenty of data cannot be structured,
which contain analysis and discussion on
enterprise-specific business matters,
leading to low information utilization.
System Integration Capability
03 Coworking software has a low frequency
of technology iteration, a difficulty in
technology selection, meanwhile the
access to data rights management has
also become a major problem.
Individualized Needs
04 According to different business content,
the individual needs of business users are
constantly overlapping, and and there is a
high demand for product flexibility.
Knowledge Graph Builds the Enterprise Brain
Enterprises can connect information dispersed in different
systems through the establishment of knowledge Graph to
form an integrated knowledge platform across
organizational structures and business domains, promoting
cross-departmental and cross-system data sharing.
NLP Techniques Extract Key Information
Enterprises can automatically analyze, categorize,
summarize and extract key knowledge from unstructured
data by using natural language processing and other
technologies to provide decision makers with effective
decision-making information.
Technological Innovation and Data Security
Personalized Systems
On the one hand, generative AI assists the innovation team
to improve creativity, and break through the iterative
innovation of new technologies; on the other hand, through
real-time monitoring of data circulation, it realizes the safe
storage and isolation of data within the enterprise.
Generative AI generates a personalized office system for the
enterprise based on the business content, past office
behaviors and preferences of enterprise users, and
continuously enhancing the application system and
functions through the feedback of the users.
❑The emergence of generative AI-related technologies has injected powerful and more direct new
functions into the digital transformation of enterprises.
•Industry Traditional Challenges: Currently, the digitalization of collaborative office is still insufficient,
which leads to the difficulty of connecting different business software and business information within
each enterprise; in addition, the extraction of key knowledge of unstructured data, the demand for
personalized office systems, and the security of information and data sharing have become the main
challenges.
•New form of "Generative AI+Enterprise Application": By leveraging knowledge graph and NLP
technologies, generative AI enables enterprises to consolidate disparate document data into a unified
cross-business and cross-departmental integration platform. This not only streamlines office operations
but also enhances decision-making quality by automatically analyzing, categorizing and extracting the
essential insights from unstructured information. Furthermore, the large model creates a personalized
office system for the enterprise based on the company's historical office behavior and preferences.
Key findings
Since the uneven digitalization of enterprises, the enterprise application industry faces challenges such
as difficulty in synchronizing office system software and weak system integration capabilities. The
emergence of generative AI can help the industry to connect internal enterprise information data based
on advanced technologies such as knowledge graphs and natural language processing, efficiently extract
effective decision-making information, and generate personalized office systems.
2.13.1 Challenges and Developments in The Enterprise
Application Industry (1/2)
Sullivan Market Research Chapter II: Compendium of Applied Practices
92
www.leadleo.com
400-072-5588
Key findings
Generative AI empowers enterprise users with its data-driven decision-making capabilities, enabling
office collaboration, and automation. It integrates the flow of internal information and key decision-
making information support, and upgrades and develops personalized products based on the analysis of
users' office behaviors, creating a new way of working.
Generative AI Opportunity Points Graph
Source: Sullivan
❑With generative AI, enterprises have the opportunity to build new generation of applications and
create new AI-driven products and offices mode.
1Office Support Software: Generative AI helps enterprise employees efficiently complete tasks such as document writing, code annotation,
etc. Employees are tuned on the basis of obtaining basic results, thus significantly improving work efficiency.
2Management System: The Agent capability assists enterprises to in automatically disassembling tasks and auto-generate work objectives,
reducing the dependence on manual execution process nodes, and realizing a systematic leap in operation and management efficiency.
3Product Application and Development: Generative AI analyzes user behavior and preferences, learning from user feedback in real time and
continuously improving its own functions and performance. In addition, it provides developers with the tools to develop and innovate
designs for product requirements.
Application and
Development
Content Generation:
Collaborative Office System
Operational Management:
Manage Customer Information
Training
support:
Product Development:
Application Feedback:
Note: Some applications involve multiple technical functions, the corresponding colors represent the main functional modules.
2.13.1 Scenario Challenges and Developments in the
Enterprise Application Industry (2/2)
Poster Material Generation
AI Review
Marketing Copy Generation
Code Comment Generation
Content Compliance Review
Office efficiency gains:
Document Information Extraction
Data Analysis and Visualization
Project Process Monitoring Alerts
E-mail Intelligent Classification
Financial Data Analysis
Optimization of Resource Allocation
Workflow Automation
Risk Monitoring and Management
Office Technology Training
Product Requirements Mining
Product Innovation Design
Code Development Assistant
Smart Product Due Diligence
User Behavior Analysis
Customer Base Data Analysis
Personalized Marketing
Reduce process delays and improve overall operational efficiency
Personalized products meet user needs
Enhance efficiency of office, promote the internal information sharing
Office Support Software Management System
Source: Sullivan
Sullivan Market Research Chapter II: Compendium of Applied Practices
Data
analysis and
forecasting
Content
generation
and design
Content
verification
and
monitoring
real time
interaction
Decision
support
Information
collection
and
organization
Value
Created
Scenarios
Opportunity points
for generative AI
applications across
scenarios in the
Enterprise
Application
industry
93
www.leadleo.com
400-072-5588
Source: Sullivan
Potential Application Risks
Source: Sullivan
Intellectual Property Protection
Copyright Infringement and Content Licensing: Generative AI generates content
through natural language modeling, which is difficult to ensure it does not involve
the copyright of existing works. Users who inadvertently use generated content for
business copywriting, or even public distribution may be involved in the risk of
plagiarism, leading to intellectual property disputes.
Copyright
Originality Assessment: Generative AI creates new content by integrating existing
text, rendering the originality of the generated content controversial. If the
generated content is not considered sufficiently original, organizations may face the
dilemma of using the content without effective IP protection.
Mandates
Secondary
Creative
License
Privacy and Data Security
Information Extraction Attack: Generative AI models may leak sensitive
information from training data. When employees use generative AI to perform
assisted work, the input information may be used as training data for further
iterations, and an attacker may infer the training data by analyzing the model's
output, leading to the disclosure of sensitive information.
Data Hegemony and Equity Issues
Interpretability and
Uncertainty Risk
Permission Control: Enterprises have high permission control needs, and data
sharing enabled by generative AI may cause ordinary employees to learn
about high-level trade secrets, which in turn bring out leakage.
Data Hegemony: Companies with access to algorithmic models with richer
resources and better data quality will gain an advantage in business and
development, creating a "technological monopoly".
Unfairness: Inadequate data representation may lead to inaccuracy in
enterprise model decision-making, especially in areas involving diversity,
where enterprises face a series of challenges ranging from model selection
and customization, in-depth mining of scenario value, to cost optimization,
arithmetic allocation, etc.
Generated Content Reliability Risk: Generative AI can output uncertain
results when faced with unseen input data, especially in critical areas such
as management and auditing, which can lead to serious losses and
consequently reduce user trust in the model.
The Problem of Model Interpretability: The complexity of generative AI
makes its decision-making mechanisms difficult to decipher, increasing the
difficulty of reviewing and validating it in applications that require
compliance with regulations, masking model bias or error.
Privilege
Control
Business
Managemen
t
Laws and
Regulations
Cost
Optimizatio
n
Audits
Functional Areas
Intellectual
Achievements
Safeguard
Arithmetic
Configuration
User
Experience
Privacy Data
Analysis
Disclosure
of
Information
Technology
Integration
Key findings
Generative AI share the risk of enterprise data and information leakage, as well as the key link
of the data lag and technical failures leading to system platforms throughout the chain of
information flow paralysis. Therefore, enterprise users and application developers need to add
supervision and the relevant level of development technology.
2.13.2 Potential Application Risks in The Enterprise
Applications Industry
Sullivan Market Research Chapter II: Compendium of Applied Practices
94
www.leadleo.com
400-072-5588
2.13.3 Best Practices in Enterprise Application Industry
- Overseas version of WPS AI
Application scenarios: Enterprise Application Industry + Office Scenarios
Core evaluation keywords
Fast and Efficient Model
Selection
R&D Efficiency
Optimization Cooperation and
Innovation
Overseas Business
Development Support
Program capacity analysis
Appraisal
scope
Functional value and applicability Technical Performance and Innovativeness
Implementation and support Experience Satisfaction Feedback
Overseas version of WPS AI
•Functional Value: Amazon Bedrock helps WPS to safely apply the
base model with its rapid deployment call, data security
guarantee, and base stability.
•Positive Benefits: Overseas version of WPS AI achieves an
improvement in R&D efficiency of 30%+ and storage cost
optimization of 40%+.
•Strategic Fit: Enhancing User Experience with Generative AI to
create a user refined operation path and extend the user scene.
•Quality Controlled: Specialized output using the Big Language
Model on Amazon Bedrock for semantic checking and document
touch-ups.
•Generate Quickly: WPS applies Amazon Bedrock to realize PPT
one-click generation, document conversion and semantic supply
to improve productivity and output quality.
•Content Compliance: Amazon Cloud Technology holds
140+ compliance tools and certifications to help the
industry through a shared responsibility model for
security compliance.
•Economic Cost: Based on Amazon S3’s intelligent
tiering feature, WPS achieves 40%+ storage cost
optimization; also,daily delayed slow down by 50%+.
•Landing support: build refined operations based on Amazon
SageMaker services, predict user purchases through artificial
intelligence, and enhance user conversion rates.
•Scenario demand satisfaction: based on
Amazon Bedrock’s rich selection of model
model , WPS can quickly select and switch
according to the needs of the actual application scene. The
WPS AI feature push can be greatly accelerated by
selecting a suitable base model, which can be completed in
two steps from selection to deployment.
•Long-term cooperation and willingness to invest: The two sides
have a long history of cooperation. In the future, WPS will
continue to cooperate with Amazon Cloud Technology in
exploring the fusion of new technologies and office software.
Sullivan Market Research Chapter II: Compendium of Applied Practices
Sources: AWS, sullivan
SPECIAL DISCLAIMER: The foregoing specific AWS Generative AI related services are only available in Amazon Web Services overseas regions.
Specific information is subject to the official website of Amazon Web Services Overseas Region (aws.amazon.com).
95
www.leadleo.com
400-072-5588
Product Structure and Core Advantages
As the global leader in cloud computing, AWS provides Amazon Bedrock, a fully managed
generative AI service for the overseas version of WPS, which enables rapid model selection and
deployment, accelerates the landing of generative AI applications, and promotes business
innovation.
Amazon Bedrock Model Claude 3 Series Models
...
One-stop intelligent office platform
Model
Option
AI21 Labs Anthropic
Meta
Cohere
Claude 3 Sonnet
Claude 3 Haiku
Core Competencies
Semantic
Information
PPT Automatic
Generation
Specialized
Output
Semantic
Censorship
Mode
Auto-matching
Document Generation
and Reworking
...
Solution Effectiveness
•Cost Reduction and Efficiency Promotion: WPS achieves petabyte-scale massive data storage through Amazon S3; realizes 40%+ storage cost
reduction and 50%+ daily response enhancement through its intelligent tiering function; and stabilizes its end-user response latency to less
than 500 milliseconds in some countries.
•Operation services: WPS based on Amazon SageMaker services to build a complete user fine-tuned operation path, extending the user scene.
•R&D Efficiency Optimization: WPS achieves R&D efficiency increase of 30%+ based on Claude 3 series model, and reaches high-speed
document modification with 1.3s-1.5s response speed.
•WPS internally needs to find a basic model that matches its
own business scenarios, and realize the advancement of WPS
AI's own functions.
•WPS externally needs to look for office solution which is more
intelligent, stable and efficient, as well as a refined and
customized operation model that can achieve user growth.
•Arithmetic: Nvidia AI chip combined with AWS's self-developed
chip empowers the WPS AI to efficiently and cost-effectively iterate
the model, realizing the planning and deployment of applications.
•Models, Frameworks and Applications: build Amazon Bedrock
model hosting services to provide overseas users with consistent
experience from text to images, and achieve all-around
empowerment.
Content
Creation
Smart
Assistant
Knowledge
Insight
Sullivan Market Research Chapter II: Compendium of Applied Practices
Effectiveness of Implementation
Client’s Demands Solutions
Sources: AWS, sullivan
SPECIAL DISCLAIMER: The foregoing specific AWS Generative AI related services are only available in Amazon Web Services overseas regions.
Specific information is subject to the official website of Amazon Web Services Overseas Region (aws.amazon.com).
AWS WPS
96
www.leadleo.com
400-072-5588
2.13.3 Best Practices in Enterprise Application Industry
- WPS AI Enterprise Edition
Application Scenario: Enterprise Application Industry + AI Office Assistant
Core evaluation keywords
Office Innovation
Programme Refined RAG Framework User Customised
Experience Full Ecological Cooperation
Model
WPS AI Enterprise Edition
Program capacity analysis
Source: Alibaba Cloud, Sullivan
Appraise
Scope
•Accurate Demand Adaptability: WPS AI Enterprise Edition
solution demonstrates high adaptability to demand, Alibaba cloud
cooperates with its client to develop AI Docs document mind and
Copilot Pro, which strongly supports the intelligentization of
Kingsoft office software functions, optimizes user experience and
improves office efficiency.
•Long-term Strategic Fit: the implementation of WPS AI Enterprise
Edition is highly aligned with Kingsoft‘s long-term strategic
direction, including multi-screen, content, and cloud,
collaboration, AI, and so on, showing its
important values in development.
Functional value and Applicability Technical Performance and Innovativeness
•Full Ecological Assistance: Alibaba Cloud combines Bailian, Tongyi
and TINGWU to provide multi-scenario functional coverage for WPS.
•Generated Content Quality: By applying the RAG framework and
Qwen-72B model to improve the accuracy, stability, and relevance
of the generated content, and apply customer feedback to product
iteration.
•Reasoning Efficiency: Alibaba Cloud design a refined solution for
Kingsoft from the inference engine, inference platform and model
structure layers.
Implementation and Support
•Deployment Cost and Time Efficiency: full-size,
multimode model provides customers with diverse,
optimized selection space. By integrating in the
customer's business code WPS eventually optimize the
internal system.
•Training and Support Services: Alibaba Cloud provides a technical
support team with rapid response capabilities, ensuring minute-
to-minute service response times and offer multiple technical
exchange to share training sessions with customers.
Customer Satisfaction Feedback
•Scenario Value Satisfaction:
WPS AI’s user interface design is
simple and intuitive, and the
operation process is smooth, so that users can quickly get
started and efficiently complete office tasks.
•Satisfaction with Experience and Customization: Alibaba Cloud
has jointly optimized and developed a number of AI functions for
WPS‘s actual needs, including content creation, rewriting and
touching up etc., to meet the needs of different users.
Sullivan Market Research Chapter II: Compendium of Applied Practices
97
www.leadleo.com
400-072-5588
Product Structure and Core Advantages - Intelligent Document Library
In view of the inefficiency of knowledge data capture, along with the complexity surrounding
office authority control, the integrated AI office solution jointly constructed by Aibaba cloud and
Kingsoft has significantly enhanced the user activicity of Kingsoft users, driven the office model
towards a cloud-based and collaborative structure, and established a novel process for intelligent
office operations.
Solution Effectiveness
Application
Layer
Kernel
Capability
Layer
Infrastructure
Services
Layer
Large Model Self-developed Large Model Tongyi-Qianwen
Intelligent Q&A Intelligent Creation …
Intelligent Extraction
Application Scenario
Application
Capability
Layer File Management Permission Management Open Capabilities
Space Management
Basic Functions
Parse Document
Segmentation
Vectorisation
Preprocess
Field Discovery
Content Extraction
LoRA
Structuring
Recalling
Merging
Sorting
RAG
OCR
Permission System
Cloud Document
Document Service
Vector Model
Sorting Model
SFT model
NLP/NLU
AI Gateway
Audit …
KPP(PE)
AI Infrastructure
ES
Redis …
MySQL
Middleware
Other Large Models
LLMs
•Variation of Knowledge: Integrate a suitable solution with
dynamic change of workflow, ensuring continuous operation
feedback throughout process of use.
•Efficient Use of Data:How to generate higher quality
intelligent creations by specifying sources and topics in
massive knowledge.
•Permission Control: Enterprises require robust mechanisms
to ensure that information is not accessed by unauthorised
individuals.
•Intelligent Assisted Creation System: By integrating intelligent
document library and AI writing, reading and analysis assistant, a
system capable of intelligent creation has been constructed to
support updating and optimization of documents.
•Refined Permission and Security Control Mechanism: A complete
document permission management system has been implemented
to achieve data isolation and secure storage.
•Improve enterprise benefits: The solution helped enterprises achieve a monthly active device count (MAD) of 602 million, a year-on-year
increase of 2.21%, Cloud document services are very popular, and office mode is moving towards cloud and collaboration. and a total
number of documents in the cloud hit 217.4 billion.
•Improve product function: The solution assisted WPS AI to go online with 20+ AI functions, and the accuracy rate of resume structured
extraction was greater than 90% in the extraction function of the online AI Docs smart document library.
Source: Alibaba Cloud, Sullivan
Sullivan Market Research Chapter II: Compendium of Applied Practices
Effectiveness of Implementation
Client’s Demands Solutions
Alibaba Cloud WPS
98
www.leadleo.com
400-072-5588
Methodology
◆Laying out the Chinese market, Leadleo Research Institute has conducted in-depth research on 19 major
industries and continuously tracked market changes in 532 vertical industries, deposited more than 1 million
industry research value data elements, and completed more than 10,000 independent research and
consulting projects.
◆Relying on China's active economic environment, Leadleo covers the entire industry development cycle.
From the establishment, development and expansion of enterprises in the industry to the maturity of listed
and post-listed enterprises, Leadleo’s researchers in various industries actively explore and evaluate the
industry's changing industrial models, business models and operation modes of enterprises, and interpret the
development of the industry with a professional vision.
◆Leadleo employs a multi-method approach to research, integrating traditional and new methodologies,
developing bespoke algorithms, and combining industry-wide big data. By utilizing a range of techniques,
we identify the underlying factors influencing quantitative data and examine the rationale behind qualitative
content. The Research Institute provides an objective and truthful description of the industry's current
situation and prospective future development trend. Its research reports present a comprehensive overview
of the industry, encompassing its past, present and future.
◆Leadleo maintains a close watch on the latest developments in the industry. The report content and data will
be updated and optimised in line with developments in the industry, technological innovation, changes in
the competitive landscape, the promulgation of policies and regulations, and market research.
◆Leadleo closely monitors the latest developments in the industry. The report content and data will be
updated and optimized in line with developments in the industry, technological innovation, changes in the
competitive landscape, the promulgation of policies and regulations, and market research.
99
www.leadleo.com
400-072-5588
Legal Notices
◆Copyright of this report belongs to the Leadleo Research Institute. No organization or individual may
reproduce, replicate, publish or quote in any form without written permission. If permission is granted to
quote, publish or otherwise use the material, it must be done within the scope of the permitted use and
indicate the source of the Leadleo Research Institute. Furthermore, any quote, abridgement or modification
must not contravene the original intent of the report.
◆The analysts of this report are fully qualified in professional research, ensuring that the data in the report is
sourced from legally compliant channels. The output of views and data analysis is based on the analysts'
objective understanding of the industry, and this report is not subject to the authorization or influence of any
third party.
◆The views or information presented in this report serve as a reference point only – no securities or fund
investment advice is provided. This report is distributed in accordance with applicable legislation and for
informational purposes only. Neither it nor any of its contents may be construed as an advertisement or a
securities research report. In accordance with applicable legislation, Leadleo is entitled to provide
investment, financing or advisory and other related services to the companies mentioned in the report.
◆Please note that this report contains information sourced from publicly available sources. Leadleo does not
guarantee the accuracy, completeness or reliability of such information. The information, opinions and
assumptions set out in this report reflect the judgement of Leadleo as at the date of issue of this report. Any
descriptions in previous reports should not be relied upon as a basis for future performance. Please notify us
promptly if you find any inconsistencies between the information, opinions and assumptions contained in
this report and any reports or articles issued Leadleo at a later date. Leadleo gives no guarantee that the
information in this report will be kept up to date. The information in this report is subject to change without
prior notice. Readers are advised to check for updates or changes. Any organization or individual using the
data, analyses, research or any part of the contents of this report shall be responsible for all activities
undertaken as a result and shall be liable for any loss or injury resulting from such activities.
©Frost & Sullivan（China）
©Leadleo
https://space.bilibili.com/647223552
https://weibo.com/u/7303360042
www.frostchina.com ; www.leadleo.com
Lead Analysts
Junjie Hu、Linye Jiang
17721418134
jackey.hu@frostchina.com
chloe.jiang@frostchina.com
Head of in-depth research team
Livia Li
13149946576
livia.li@frostchina.com
